@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix skos: . vivo:departmentOrSchool "Applied Science, Faculty of"@en, "Civil Engineering, Department of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Essa, Mohamed"@en ; dcterms:issued "2015-10-24T04:59:40"@en, "2015"@en ; vivo:relatedDegree "Master of Applied Science - MASc"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description "Recently, there has been a growing interest in using microsimulation models for the safety assessment of road facilities by analyzing vehicle trajectories and estimating conflict indicators. Using microsimulation in safety studies can have several advantages. However, concerns have been raised about the ability of these models to realistically represent unsafe vehicle interactions and near misses and the need for a rigorous model calibration. The main objective of this thesis is to investigate the relationship between field-measured traffic conflicts and simulated traffic conflicts at signalized intersections. Automated video-based computer vision techniques were used to extract vehicle trajectories and identify field-measured rear-end conflicts. Conflict measures (e.g. time-to-collision (TTC)) and locations were determined and compared with simulated conflicts from the Surrogate Safety Assessment Model (SSAM) by analyzing the vehicles trajectories extracted from two microsimulation models: VISSIM and PARAMICS. To increase the correlation between simulated and field-measured conflicts, a two-step calibration procedure of the simulation models was proposed and validated. In the first calibration step, the simulation model was calibrated to ensure that the simulation gives reasonable results of average delay times. Then, in the second calibration step, a Genetic Algorithm procedure was used to calibrate the safety-related parameters in the simulation model. The correlation between simulated and field-measured conflicts was investigated at different thresholds of TTC. The results obtained from VISSIM and PARAMICS were compared. Furthermore, the transferability of the calibrated simulation models for safety analysis between different sites was investigated. As well, the spatial distributions of the field-measured and the simulated conflicts were compared through conflict heat maps. Overall, good correlation between field-measured and simulated conflicts was obtained after calibration for both models especially at higher TTC values. Also, the results showed that the simulation model parameters are generally transferable between different locations as the transferred parameters provided better correlation between simulated and field-measured conflicts than using the default parameters. The heat maps showed that there were major differences between field-measured and simulated conflicts spatial distribution for both simulation models. This indicates that despite the good correlation obtained, both PARAMICS and VISSIM do not capture the actual conflict occurrence mechanism."@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/54854?expand=metadata"@en ; skos:note "CALIBRATION AND VALIDATION OF TRAFFIC MICROSIMULATION MODELS FOR SAFETY EVALUATION USING AUTOMATED VIDEO-BASED CONFLICT ANALYSIS by Mohamed Essa B.Sc., Ain Shams University, 2008 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Civil Engineering) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) September 2015 © Mohamed Essa, 2015 ii Abstract Recently, there has been a growing interest in using microsimulation models for the safety assessment of road facilities by analyzing vehicle trajectories and estimating conflict indicators. Using microsimulation in safety studies can have several advantages. However, concerns have been raised about the ability of these models to realistically represent unsafe vehicle interactions and near misses and the need for a rigorous model calibration. The main objective of this thesis is to investigate the relationship between field-measured traffic conflicts and simulated traffic conflicts at signalized intersections. Automated video-based computer vision techniques were used to extract vehicle trajectories and identify field-measured rear-end conflicts. Conflict measures (e.g. time-to-collision (TTC)) and locations were determined and compared with simulated conflicts from the Surrogate Safety Assessment Model (SSAM) by analyzing the vehicles trajectories extracted from two microsimulation models: VISSIM and PARAMICS. To increase the correlation between simulated and field-measured conflicts, a two-step calibration procedure of the simulation models was proposed and validated. In the first calibration step, the simulation model was calibrated to ensure that the simulation gives reasonable results of average delay times. Then, in the second calibration step, a Genetic Algorithm procedure was used to calibrate the safety-related parameters in the simulation model. The correlation between simulated and field-measured conflicts was investigated at different thresholds of TTC. The results obtained from VISSIM and PARAMICS were compared. Furthermore, the transferability of the calibrated simulation models for safety analysis between different sites was investigated. As well, the spatial distributions of the field-measured and the simulated conflicts were compared through conflict heat maps. Overall, good correlation between field-measured and simulated conflicts was obtained after calibration for both models especially at higher TTC values. Also, the results showed that the simulation model parameters are generally transferable between different locations as the transferred parameters provided better correlation between simulated and field-measured conflicts than using the default parameters. The heat maps showed that there were major differences between field-measured and simulated conflicts spatial distribution for both simulation models. This indicates that despite the good correlation obtained, both PARAMICS and VISSIM do not capture the actual conflict occurrence mechanism. iii Preface Portions of the introductory text in Chapter 1, portions of the literature review in Chapter 2, portions of field-measured data description in Chapter 3, and a version of Chapter 6 have been published [Essa, Mohamed, and Tarek Sayed. “Transferability of calibrated microsimulation model parameters for safety assessment using simulated conflicts”, Accident Analysis & Prevention, Volume 84, November 2015, Pages 41-53, ISSN 0001-4575]. I conducted all the analysis and wrote most of the manuscript. Portions of the introductory text in Chapter 1, portions of the literature review in Chapter 2, portions of field-measured data description in Chapter 3, and a version of Chapter 4 have been accepted for publication [Essa, Mohamed, and Tarek Sayed. \"Simulated Traffic Conflicts: Do They Accurately Represent Field-Measured Conflicts?\" Transportation Research Board 94th Annual Meeting. No. 15-0707. 2015. Accepted for publication in Transportation Research Record, 2015 (In Press)]. I conducted all the analysis and wrote most of the manuscript. The analysis of data in Section 3.2 in Chapter 3 is based on the work done in the transportation research office at the University of British Columbia by Tageldin et al. [Tageldin, Ahmed, Tarek Sayed, Mohamed Zaki, and Mohamed Azab. \"A Safety Evaluation of an Adaptive Traffic Signal Control System using Computer Vision.\" Advances in Transportation Studies 2, no. Special Issue (2014): 83-96]. iv Table of Contents Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iii Table of Contents ......................................................................................................................... iv List of Tables ................................................................................................................................ ix List of Figures ............................................................................................................................... xi List of Abbreviations ................................................................................................................. xiv Acknowledgements ......................................................................................................................xv Dedication ................................................................................................................................... xvi Chapter 1: Introduction ................................................................................................................1 1.1 Background ..................................................................................................................... 1 1.1.1 Traditional Traffic Safety Analysis ............................................................................ 2 1.1.2 Traffic Conflict Technique ......................................................................................... 2 1.1.3 Automated Video-based Computer Vision Techniques ............................................. 2 1.1.4 Surrogate Safety Assessment Model (SSAM) ............................................................ 3 1.2 Research Objective ......................................................................................................... 4 1.3 Thesis Structure .............................................................................................................. 7 Chapter 2: Literature Review .......................................................................................................9 2.1 Traditional Traffic Safety Analysis ................................................................................ 9 2.2 Traffic Conflict Technique (TCT) ................................................................................ 10 2.2.1 Safety Continuum and Traffic Conflict Hierarchy ................................................... 11 2.2.2 Traffic Conflict Indicators ........................................................................................ 13 2.2.2.1 Time to Collision (TTC) ................................................................................... 13 v 2.2.2.2 Post-encroachment Time (PET) ........................................................................ 16 2.2.2.3 Gap Time (GT) ................................................................................................. 16 2.2.2.4 Deceleration-to-safety Time (DST) .................................................................. 17 2.2.3 Challenges to Traffic Conflict Technique ................................................................. 17 2.2.3.1 Consistency in Conflict Definition ................................................................... 17 2.2.3.2 Validity of TCT................................................................................................. 18 2.2.3.3 Reliability in Conflict Measurements ............................................................... 19 2.3 Automated Video-based Computer Vision Techniques ............................................... 20 2.3.1 Tracking of Road Users ............................................................................................ 21 2.3.1.1 3D Model-based Tracking ................................................................................ 21 2.3.1.2 Region-based Tracking ..................................................................................... 21 2.3.1.3 Contour-based Tracking.................................................................................... 22 2.3.1.4 Feature-based Tracking ..................................................................................... 22 2.3.2 Computer Vision at UBC .......................................................................................... 22 2.4 Traffic Microsimulation Models and SSAM ................................................................ 24 2.4.1 Surrogate Safety Assessment Model (SSAM) .......................................................... 24 2.4.2 Calibration of Microsimulation Models.................................................................... 25 Chapter 3: Field-Measured Data ................................................................................................29 3.1 Study Location and Data Collection ............................................................................. 29 3.1.1 The First Intersection ................................................................................................ 29 3.1.2 The Second Intersection ............................................................................................ 32 3.2 Automated Video-Based Computer Vision Analysis ................................................... 33 3.2.1 Video Encoding ........................................................................................................ 34 vi 3.2.2 Camera Calibration ................................................................................................... 35 3.2.2.1 Corresponding Points ........................................................................................ 36 3.2.2.2 Distances ........................................................................................................... 36 3.2.2.3 Angles ............................................................................................................... 37 3.2.2.4 Global Up Directions ........................................................................................ 37 3.2.3 Calibration Verification ............................................................................................ 37 3.2.4 Feature Tracking ....................................................................................................... 37 3.2.5 Feature Grouping (Objects Creation) ........................................................................ 38 3.2.6 Prototype Generation and Matching ......................................................................... 39 3.2.7 Conflicts Analysis ..................................................................................................... 39 Chapter 4: The Microsimulation Model VISSIM .....................................................................41 4.1 VISSIM Calibration Framework................................................................................... 41 4.2 VISSIM Platform .......................................................................................................... 42 4.3 The VISSIM Calibration Procedure .............................................................................. 44 4.3.1 First Step Calibration ................................................................................................ 45 4.3.2 Second Step Calibration ............................................................................................ 48 4.3.2.1 Sensitivity Analysis .......................................................................................... 49 4.3.2.2 Genetic Algorithm ............................................................................................ 52 4.3.3 The Calibration Results............................................................................................. 54 4.4 Validation of the Proposed Procedure .......................................................................... 56 4.5 Conflicts Spatial Distribution ....................................................................................... 59 4.6 Summary ....................................................................................................................... 63 Chapter 5: PARAMICS – VISSIM Comparison ......................................................................64 vii 5.1 PARAMICS Calibration Framework ............................................................................ 64 5.2 PARAMICS Platform ................................................................................................... 64 5.3 The PARAMICS Calibration Procedure ....................................................................... 67 5.3.1 First Step Calibration ................................................................................................ 68 5.3.2 Second Step Calibration ............................................................................................ 70 5.3.3 The Calibration Results............................................................................................. 72 5.4 Validation of the Proposed Procedure .......................................................................... 74 5.5 PARAMICS - VISSIM Comparison ............................................................................. 77 5.5.1 Car-Following-Model and Safety-Related Parameters ............................................. 78 5.5.2 Correlation between Simulated and Field-measured Conflicts ................................ 80 5.5.3 Conflict Spatial Distribution ..................................................................................... 84 5.6 Summary ....................................................................................................................... 86 Chapter 6: Transferability of Simulation Models.....................................................................89 6.1 The Need of Transferability Investigation .................................................................... 89 6.2 Transferability Approaches ........................................................................................... 90 6.3 Simulation Model Transferability Investigation ........................................................... 90 6.3.1 Scenario 1.................................................................................................................. 92 6.3.2 Scenario 2.................................................................................................................. 92 6.3.3 Scenario 3.................................................................................................................. 92 6.3.4 Scenario 4.................................................................................................................. 93 6.3.5 Scenario 5.................................................................................................................. 93 6.4 Transferability of Individual Parameters ...................................................................... 95 6.5 Conflict Spatial Distribution ......................................................................................... 96 viii 6.6 Summary ....................................................................................................................... 99 Chapter 7: Summary and Conclusions ....................................................................................101 7.1 Summary ..................................................................................................................... 101 7.2 Conclusions ................................................................................................................. 102 7.3 Recommendations for Future Research ...................................................................... 104 Bibliography ...............................................................................................................................106 Appendices ..................................................................................................................................118 Appendix A Traffic Data Details ............................................................................................ 118 A.1 Traffic Volumes ...................................................................................................... 118 A.2 Platoon Ratios and Delays ...................................................................................... 122 A.3 Traffic Signal Programs .......................................................................................... 124 Appendix B Field-Measured Conflict Data ............................................................................ 125 Appendix C Sensitivity Analysis ............................................................................................ 128 Appendix D Computer Codes ................................................................................................. 132 D.1 MATLAB Script (Genetic Algorithm for VISSIM) ............................................... 132 D.2 MATLAB Script (Editing of the VISSIM (*inp) File) ........................................... 135 D.3 VISUAL BASIC Code (Automatic Runs of VISSIM Simulation) ........................ 136 D.4 MATLAB Script (Genetic Algorithm for PARAMICS) ........................................ 137 D.5 MATLAB Script (Editing of the PARAMICS Configuration File and Run Simulation) .......................................................................................................................... 140 D.6 MATLAB Function (Check Traffic Signal Indications) ........................................ 140 ix List of Tables Table 4.1 Sensitivity analysis results ............................................................................................ 50 Table 4.2 Description of the selected VISSIM model parameters................................................ 51 Table 5.1 Description of the selected PARAMICS model parameters ......................................... 71 Table 5.2 PARAMICS and VISSIM model parameters ............................................................... 79 Table 6.1 Five scenarios at the second intersection for transferability investigation ................... 91 Table 6.2 Calibrated VISSIM parameters from the first and the second intersections ................. 96 Table A.1 Traffic volume of the first intersection (west approach) ........................................... 118 Table A.2 Traffic volume of the first intersection (east approach)............................................. 118 Table A.3 Traffic volume of the first intersection (north approach) .......................................... 119 Table A.4 Traffic volume of the first intersection (south approach) .......................................... 119 Table A.5 Traffic volume of the second intersection (west approach) ....................................... 120 Table A.6 Traffic volume of the second intersection (east approach) ........................................ 120 Table A.7 Traffic volume of the second intersection (north approach) ...................................... 121 Table A.8 Traffic volume of the second intersection (south approach) ..................................... 121 Table A.9 Platoon ratios and delays of the first intersection (east and west approaches) .......... 122 Table A.10 Platoon ratios and delays of the first intersection (north and south approaches) ..... 122 Table A.11 Platoon ratios and delays of the second intersection (east and west approaches) ... 123 Table A.12 Traffic signal program of the first intersection ........................................................ 124 Table A.13 Traffic signal program of the second intersection ................................................... 124 Table B.1 Field-measured traffic conflicts of the first intersection (calibration dataset) ........... 125 Table B.2 Field-measured traffic conflicts of the first intersection (validation dataset) ........... 126 Table B.3 Field-measured traffic conflicts of the second intersection ....................................... 127 x Table C.1 Sensitivity analysis of the car following model parameters....................................... 128 Table C.2 Sensitivity analysis of the car following model parameters (Continued) .................. 129 Table C.3 Sensitivity analysis of the lane change parameters .................................................... 130 Table C.4 Sensitivity analysis of the signal control parameters and the desired deceleration ... 131 xi List of Figures Figure 2.1 The safety pyramid (Source: Hydén, 1987) ................................................................ 12 Figure 2.2 Example of possible diamond hierarchy shape (Source: Svensson, 1998) ................. 13 Figure 3.1 Study location .............................................................................................................. 29 Figure 3.2 Cameras positions at the first intersection ................................................................... 31 Figure 3.3 Cumulative distribution curve of desired speed .......................................................... 32 Figure 3.4 Cameras positions at the second intersection .............................................................. 33 Figure 3.5 A schematic diagram of the process of video analysis ................................................ 34 Figure 3.6 Four types of annotations used for camera calibration process (Source: Tageldin et al., 2014) ............................................................................................................................................. 36 Figure 4.1 General framework of the simulation model calibration process ................................ 42 Figure 4.2 VISSIM model for the first intersection ...................................................................... 44 Figure 4.3 Arrival type calibration process (VISSIM) ................................................................. 46 Figure 4.4 Arrival types for calibration dataset (VISSIM) ........................................................... 47 Figure 4.5 Average delays for calibration dataset (VISSIM) ....................................................... 48 Figure 4.6 Genetic Algorithm procedure flow chart ..................................................................... 53 Figure 4.7 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (VISSIM) ........................................................................................................ 55 Figure 4.8 Correlation between simulated and field conflict rates at different TTC thresholds for calibration dataset (VISSIM) ........................................................................................................ 56 Figure 4.9 Arrival types for validation dataset (VISSIM) ............................................................ 57 Figure 4.10 Average delays for validation dataset (VISSIM) ...................................................... 58 xii Figure 4.11 Correlation between simulated and field conflicts at different TTC thresholds for validation dataset (VISSIM) ......................................................................................................... 59 Figure 4.12 Heat maps of calibration dataset (VISSIM) .............................................................. 61 Figure 4.13 Heat maps of validation dataset (VISSIM) ............................................................... 62 Figure 5.1 Screen shot of the PARAMICS model for the first intersection ................................. 67 Figure 5.2 Arrival type calibration process (PARAMICS) .......................................................... 68 Figure 5.3 Arrival types for calibration dataset (PARAMICS) .................................................... 69 Figure 5.4 Average delays for calibration dataset (PARAMICS) ................................................ 70 Figure 5.5 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (PARAMICS) ................................................................................................. 73 Figure 5.6 Arrival types for validation dataset (PARAMICS) ..................................................... 75 Figure 5.7 Average delays for validation dataset (PARAMICS) ................................................. 75 Figure 5.8 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (PARAMICS) ................................................................................................. 77 Figure 5.9 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (VISSIM - PARAMICS) ................................................................................ 81 Figure 5.10 Correlation between simulated and field conflicts at different TTC thresholds for validation dataset (VISSIM - PARAMICS) ................................................................................. 82 Figure 5.11 Regression models using simulated and field-measured conflicts ............................ 83 Figure 5.12 Heat maps of calibration dataset (Field conflicts – PARAMICS conflicts – VISSIM conflicts) ....................................................................................................................................... 85 Figure 5.13 Heat maps of validation dataset (Field conflicts – PARAMICS conflicts – VISSIM conflicts) ....................................................................................................................................... 86 xiii Figure 6.1 Correlation between simulated and field-measured conflicts/conflict rates at the second intersection for scenarios (1), (2), (3), (4) and (5) ............................................................ 94 Figure 6.2 Spatial distribution of the field-measured conflicts, simulated conflicts with transferred parameters (Scenario 4), and simulated conflicts with calibrated parameters (Scenario 5) ................................................................................................................................................... 98 xiv List of Abbreviations ASC Actuated Signal Controller DeltaS Speed Differential DR Deceleration Rate DRAC Deceleration rate required to avoid a crash DST Deceleration to Safety Distance FHWA Federal Highway Administration GA Genetic Algorithm GT Gap Time LCSS Longest Common Sub Sequence LHS Latin Hypercube Sampling LOS Level of Service MADR Maximum Available Deceleration Rate MAPE Mean Absolute Percent Error MaxS Maximum Speed MRT Mean driver reaction time MTH Mean target headway NGSIM Next Generation Simulation PARAMICS Parallel Microscopic Simulation (Traffic Microsimulation Model) PET Post-encroachment Time RBC Ring Barrier Controller SSAM Surrogate Safety Assessment Model TCT Traffic Conflict Technique TET Time-Exposed-TTC TIT Time-Integrated-TTC TTA Tme To Accident TTC Time To Collision UBC University of British Columbia VISSIM Verkehr In Städten - SIMulationsmodell (German for \"Traffic in cities - simulation model\") (Traffic Microsimulation Model) WHO World Health Organization xv Acknowledgements I would like to express my deepest gratitude to my advisor, Dr. Tarek Sayed, for his excellent guidance, caring, patience, and providing me with an excellent atmosphere for doing research. I would like to thank my colleagues from the transportation group, in special, to Dr. Mohamed Zaki, Mohamed Hussein, and Ahmed Tageldin for their friendship and help in several aspects of my thesis. I am thankful for Jarvis Autey for proofreading parts of this thesis. I would also like to thank Dr. Mohamed El Esawey for encouraging and advising me many times during the past three years. I would also like to thank my parents. They were always supporting me and encouraging me with their best wishes. Finally, I would like to thank my wife, Passant. She was always there cheering me up and stood by me through the good and bad times. xvi Dedication To My Parents1 Chapter 1: Introduction This chapter contains four sections. The first section presents background information on traffic safety analysis including: traditional traffic safety analysis, traffic conflict techniques, automated video-based computer vision techniques, and the surrogate safety assessment model (SSAM). The second section discusses the research objectives. The third section describes the thesis structure. 1.1 Background The first documented traffic fatality in 1896 shed light on the importance of road safety as one of the main goals of transportation. According to the World Health Organization (WHO), 1.2 million people are killed and up to 50 million are injured on roads every year (World Health Organization, 2004). According to the Canadian Motor Vehicle Traffic Collision Statistics, 1,923 fatalities and 10,315 injuries were registered on Canadian roads in 2013. Within the period from 1994 to 2013, the total number of registered fatalities was 52,171, and the total number of registered injuries was 4,057,684 (Transport Canada, 2013). According to Transport Canada (2011), “Canada has made significant progress in reducing the number and rate of fatalities and serious injuries over the past decade compared to the period from 1996 to 2001, in part due to the success of a wide variety of road safety measures that have been implemented over these years”. Despite this progress, traffic collisions still constitute one of the main causes of injuries and fatalities. The challenge is to move toward zero fatalities and serious injury collisions (Transport Canada, Road Safety in Canada, 2011). In order to make roads safer, transportation practitioners should know how to estimate the safety impacts of different options of planning, design, and traffic management. Therefore, traffic safety analysis is crucial to evaluate the safety performance of various road facilities. 2 Traffic safety analysis can include: identification of the hazardous locations, evaluation of safety programs, verifying the effectiveness of the countermeasures applied to improve traffic safety, and improving road user behaviour. 1.1.1 Traditional Traffic Safety Analysis Traditional traffic safety analysis, which usually relies on the use of historical collision data, is a reactive approach that offers less complete understanding of collision mechanisms and how safety measures work (Sayed and Zein, 1999; Sayed et al., 1994; Autey et al., 2012). The traditional approach is associated with several shortcomings. These include: poor collision data quality and availability, and an ethical dilemma as collisions, that are required to be prevented, have to occur and be recorded over an adequately long period in order to conduct a sound safety diagnosis (Sayed and Zein, 1999; Chin et al., 1997; Ismail et al., 2010). 1.1.2 Traffic Conflict Technique Due to the limitations associated with using historical collision data in safety analysis, there has been a growing interest in traffic safety analysis techniques that rely on other surrogate safety measures. One commonly used surrogate measures are traffic conflicts or near misses. The traffic conflict technique (TCT) involves recording and evaluating the frequency and severity of near misses at a location which enables the safety professionals to immediately observe unsafe driving maneuvers at road locations without waiting for collisions to occur (Amundsen and Hydén, 1977; Svensson et al., 2006). However, the TCT is commonly criticized for the high cost of data collection and the reliability of the manual collection of the conflict data. 1.1.3 Automated Video-based Computer Vision Techniques Recently, video-based automated computer vision techniques were shown to be useful for automated traffic conflict detection and in conducting various safety analysis applications such 3 as before and after safety evaluations and studying road user behavior (Autey et al., 2012; Sayed et al, 2012; Sayed et al., 2013). The video-based computer vision approach can have many advantages. First, automated video-based analysis overcomes the shortcomings associated with the manual collection of conflict data in terms of the cost and the data reliability. Second, automated analysis of field conflicts depends on tracking vehicles trajectories; and therefore it can have more accuracy in terms of determining conflict severity (e.g. TTC) and conflict location. Third, video cameras have the ability of collecting rich and detailed traffic data. Fourth, the installation of video cameras is easier than the installation of the other sensors such as the magnetic loop detectors which requires the road surface to be excavated. 1.1.4 Surrogate Safety Assessment Model (SSAM) The use of traffic simulation models for conducting conflict-based safety evaluations has also been recently proposed (Gettman et al., 2003). There is a growing interest in using simulation models for the safety assessment of road facilities by analyzing simulated vehicle trajectories and estimating conflict indicators. The main advantages of this approach are: 1) the ability to evaluate the safety of various design and traffic management options of road facilities before actually making changes and 2) the ease of estimating simulated conflicts without actually observing them. The Surrogate Safety Assessment Model (SSAM) was recently proposed by (Siemens Energy & Automation, Inc.) and was sponsored by the Federal Highway Administration (FHWA) of the United States. The SSAM was developed to estimate traffic conflicts using simulated vehicle trajectories exported from four commonly-used microscopic simulation models: VISSIM, AIMSUN, PARAMICS, and TEXAS. Several traffic conflict indicators can be calculated including the time to collision (TTC), post-encroachment time (PET), deceleration rate (DR), maximum speed (MaxS), and speed differential (DeltaS). In 4 SSAM, conflicts are classified into three maneuver types: rear-end; lane-change; and crossing. Conflicts are identified based on conflict angle and specific thresholds for TTC and PET, which are predetermined by the user (Gettman et al., 2008). 1.2 Research Objective The SSAM approach has seen increasing use in safety evaluations recently (Shahdah et al., 2014; Wang et al., 2014; among others). However, several researchers raised two main concerns about using simulated conflicts in safety studies (Tarko and Songchitruksa, 2005; Saunier et al., 2007). First, vehicles in the simulation models follow specific rules which are aimed at avoiding collisions. Therefore, it is very difficult to represent unsafe vehicle interactions and near misses. Second, there are many model parameters in micro-simulation models. These parameters can have a significant impact on the estimated simulated conflicts. As well, there are usually several ways to model traffic in simulation models (road geometry, priority rules, conflicts areas, traffic distribution, etc.). Thus, the results can vary significantly depending on the approach used in modeling. Therefore, there is a need to investigate the relationship between simulated and field-measured conflicts before depending on the simulation models in safety evaluation. Many previous studies investigated the relationship between field and simulated conflicts such as (Huang et al., 2013; Fan et al., 2013). However, in most of them, field conflicts were collected manually and compared to simulated conflicts. Only few studies use real vehicles trajectories to automatically identify field conflicts instead of the manually collection. Automated analysis of field conflicts can have more accuracy in terms of conflict severity (e.g. TTC) and conflict location. Also, automated vehicle trajectories can help in estimating traffic characteristics such as volume, speed, and travel time. Therefore, the first objective of this study is to investigate the relationship between field-measured and simulated conflicts using automated 5 video-based computer-vision techniques to extract vehicle trajectories and identify field rear-end conflicts. Conflict measures (e.g. TTC) and location were determined and compared with simulated conflicts from VISSIM using the SSAM tool at two urban signalized intersections. By knowing conflict severity and location from the automated conflict analysis, the relationship between simulated and field-measured conflicts can be investigated at different thresholds of TTC. Also, field-measured and simulated conflict locations can be compared using heat maps. The results show that the default simulation model parameters give poor correlation between simulated and field-measured conflicts. Many previous studies indicated that traffic simulation models need to be well calibrated to give reasonable results. Previous studies such as (Cunto and Saccomanno, 2008; Huang et al., 2013) presented some calibration procedures to calibrate simulation models for safety evaluation at signalized intersections. Usually, the calibration procedure is an optimization process which aims to obtain the best values of the most important parameters in the simulation model. However, the calibration of the simulation models to produce the real traffic arrival type at signalized intersections has generally been ignored by the researchers. The arrival type describes the quality of the progression which is a critical characteristic that can be quantified by the platoon ratio. The platoon ratio depends on the proportion of all vehicles arriving during green and the ratio of effective green time to the cycle length (Highway Capacity Manual, 2000). The traffic arrival type, at signalized intersections, can impact the opportunity for interacting between vehicles; and therefore impact the number of conflicts. Therefore, the second objective of this study is to propose and validate a simplified two-step calibration procedure of the VISSIM simulation model to increase the correlation between simulated and field conflicts at signalized intersections. In the first calibration step, the arrival 6 type and the desired speed are calibrated to match the field conditions, and to ensure that the simulation model gives reasonable results of average delay times. In the second calibration step, sensitivity analysis followed by a Genetic Algorithm (GA) procedure are applied to calibrate the safety related parameters in VISSIM to enhance the correlation between simulated and field measured conflicts. The SSAM approach can be used with trajectories extracted from several traffic microsimulation models. The traffic simulation models have different car following models, and therefore they give different extracted vehicle trajectories and different frequency of simulated conflicts. The third objective of this study is to investigate the relationship between field-measured and simulated conflicts at signalized intersections using another microsimulation model (PARAMICS). Also, the applicability of the two-step calibration procedure applied to VISSIM will be investigated and validated using PARAMICS. The fourth objective of this study is to compare the results obtained from VISSIM with those obtained from PARAMICS. The comparison will include three aspects: 1) car-following model and safety-related parameters; 2) correlation between simulated and field-measured conflicts; and 3) conflict spatial distribution. Practically, the calibration process can be relatively complicated, time consuming, and need real traffic conflict data to be collected. In addition, the main advantage of using the traffic simulation models and SSAM is the evaluation of the safety of changes to design and traffic control that have not been implemented. Therefore, the transfer of the calibrated parameters of simulation models for use in safety studies between different locations (with comparable traffic conditions and geometric characteristics) can help in eliminating or reducing the need for the complex calibration process. As such, there is a need to investigate and confirm the 7 transferability of the calibrated simulation model parameters for use at different sites. Therefore, the fifth objective of this study is to investigate the transferability of calibrated parameters of microscopic simulation models for safety analysis between two different sites. The study investigates whether the transferred parameters, when applied to other sites, give reasonable results in terms of the relationship (correlation) between the field-measured and the simulated rear-end conflicts. In summary, the main objectives of this thesis can be summarized as following: 1. Using microsimulation models and automated video-based analysis to investigate the relationship between simulated and field-measured conflicts at signalized intersections. 2. To propose and validate a simplified two-step calibration procedure of VISSIM to enhance the correlation between simulated and field-measured conflicts. 3. To investigate the applicability of the proposed calibration procedure to another microsimulation model (PARAMICS). 4. To compare VISSIM and PARAMICS simulation models in terms of the car-following model and safety-related parameters; correlation between simulated and field-measured conflicts; and conflict spatial distribution. 5. To investigate the transferability of the calibrated parameters of microscopic simulation models for safety analysis at other sites. 1.3 Thesis Structure This thesis is divided into seven chapters. The first chapter provides an introduction to the thesis by presenting background information; the research objectives; and the thesis structure. The second chapter presents literature review of the previous work related to this study. The third chapter provides a detailed description of the field-measured data and the automated video-based 8 computer vision technique used in estimating traffic conflicts. The fourth chapter presents a detailed explanation of the proposed two-steps procedure of calibration of the simulation model using the microsimulation model VISSIM. The fifth chapter presents the applicability of the proposed procedure using another microsimulation model (PARAMICS), and provides a comparison between the two microsimulation models: VISSIM and PARAMICS. The sixth chapter presents the transferability of the simulation models between different sites for safety evaluation. The seventh chapter contains the research summary, conclusions, discussions, and suggestions for future research. 9 Chapter 2: Literature Review This chapter presents an overview of the subject areas related to the research topics presented in this thesis to provide the current state of the art on which this research is based. Four sections covering four topics are presented in this chapter. The first section discusses the traditional traffic safety analysis approach and the shortcoming associated with this approach. The second section provides a review of the traffic conflict technique. This includes conflict hierarchy, conflict indicators, and the challenges facing this technique. The third section is related to the development of the computer video analysis and provides information on the computer vision tool developed at the University of British Columbia (UBC). The fourth section reviews previous key studies related to the use of microsimulation models and the Surrogate Safety Assessment Model (SSAM) in safety evaluation. As a considerably high number of previous studies related to traffic conflict technique and traffic microsimulation models was found; the literature review in this chapter focuses on a reduced list of selected key studies that are highly cited. 2.1 Traditional Traffic Safety Analysis In traditional traffic safety analysis, road safety analysts typically use accident data to identify hazard locations, to diagnose safety problems, to provide options to improve safety, and to evaluate the economic feasibility of safety improvements. This is a reactive, collision-based, approach which relies on the use of historical collision data drawn mainly from collision records, police reports, and insurance claims (de Leur and Sayed, 2001; de Leur and Sayed, 2003). The traditional approach is associated with several shortcomings. Firstly, it is necessary to have a sufficient database of collision records to obtain statistically sound results. Such database requires a long time period and a large number of collisions to be recorded. As such, this approach is not valid for short term safety assessment. Second, the collision data in most 10 cases has problems represented in its unavailability and poor quality. These problems are found in the literature such as low reported rates, incomplete and miss-reported information, data entry errors, and statistical challenges due to the fact that the collisions are rare and random events (Davis, 2004; Hauer, 2006; Hirst et al., 2004; Nicholson, 1985; Saunier and Sayed, 2007). Third, there is an ethical dilemma associated with this approach as collisions have to occur and be recorded over an adequately long period in order to conduct a sound safety diagnosis. Therefore, this collision-based safety analysis approach is hampered by several limitations, and criticized as reactive and unethical by many researchers (e.g., Sayed and Zein, 1999; Ismail et al., 2010; Chin and Quek, 1997; Debnath and Chin, 2006; Songchitruksa and Tarko, 2006). 2.2 Traffic Conflict Technique (TCT) Due to the limitations associated with using historical collision data in safety analysis, there has been a growing interest in traffic safety analysis techniques that rely on other surrogate safety measures. To be acceptable, generally any surrogate measure must meet two conditions. First, it must be correlated with the clinically meaningful outcome. Second, it must fully capture the effect of the treatment (Tarko et al., 2009). Surrogate measures of traffic safety must be more frequent than collisions, have a statistical relationship to collisions, and have the characteristics of near-accidents in a hierarchal continuum from collisions to safe, undisturbed passages (Archer, 2004). One commonly used surrogate measure is traffic conflicts or near-misses. The traffic conflict technique (TCT) involves recording and evaluating the frequency and severity of near misses at a location which enables the safety professionals to immediately observe unsafe driving maneuvers at road locations without waiting for collisions to occur (Amundsen and Hydén, 1977; Svensson et al., 2006). The concept of traffic conflicts was first proposed by 11 Perkins and Harris (1968) as an alternative to collision data. Their objective was to identify traffic events that occur frequently and can be related to traffic collisions. In their study, Perkins and Harris observed and counted instances in which drivers took evasive actions to avoid being involved in collisions. Such actions which presuppose the presence of critical situations, are to be identified by some observable responses made by drivers such as a sudden changing of lanes or hard braking evidenced by the appearance of the brake lights. This approach came to be called the traffic conflict technique (Perkins and Harris, 1968; Chin and Quek, 1997). Traffic conflict is defined as “an observable situation in which two or more road users approach each other in space and time to such an extent that there is a risk of collision if their movements remained unchanged” (Amundsen and Hydén, 1977). The use of traffic conflicts for safety diagnosis has been gaining acceptance among road safety researchers as a surrogate or a complementary approach to the collision data analysis approach for many reasons. First, traffic conflicts are more frequent than road collisions and are of marginal social cost. Second, Traffic conflicts provide insight into the failure mechanism that leads to road collisions. Third, conflict data can be gathered within a shorter time period compared to accident data, and the analysis will be less affected by time-dependent factors. Also, the ethical problem associated with the need of a long collision history will be solved. Fourth, the effectiveness of any safety program can also be assessed in a shorter period of time (Chin and Quek, 1997). Thus, the traffic conflict technique addresses several shortcomings associated with collision data, including data frequency and data quality. 2.2.1 Safety Continuum and Traffic Conflict Hierarchy The interaction between road users can be hypothesized as a series of time-dependent events that range from undisturbed passages to actual collisions. This is referred to as the safety continuum 12 of traffic events (Archer, 2005; Hydén, 1987; van der Horst, 1990). The theory of severity hierarchy of traffic events was firstly proposed by Hydén (1987). A visual representation of the frequency and severity of these events was proposed as a pyramid as shown in Figure 2.1. The tip of the pyramid represents actual collisions and the base represents undisturbed passage, while the cross-sectional area of the pyramid represents the frequency of that severity level of interactions. It is well recognized that collisions are observed to occur less frequently than undisturbed passage. Therefore, the pyramid shape is logical. Although the extremes of the pyramid can be firmly detected (collisions and undisturbed passage), intermediate events still need clear definitions, thresholds, and effective measurement procedures. Figure 2.1 The safety pyramid (Source: Hydén, 1987) According to the safety pyramid proposed by Hydén (1987), there would be a continuous increase in the number of events as the severity of the event decreases. This is obviously true when all events are considered. However, if only interactions characterised by the presence of a collision course are considered, then it is likely that the shape is different (Svensson and Hydén, 2006). Instead of the pyramid shape, Svensson (1998) proposed a diamond shape for safety heirarchy as shown in Figure 2.2. In fact, this is a more logical shape as the number of interactions does not continously increase when the severity of the interaction decreases. There 13 should be a lowest limit of severity at which there is a probability of taking an evasive action. Below this limit, the probability of taking evasive action must be zero. Therefore, a bottom with a peaked shape (diamod shape) must be more realistic than a bottom with a straight line in the traditional pyramid which there are numerous interactions (Svensson, 1998). Figure 2.2 Example of possible diamond hierarchy shape (Source: Svensson, 1998) 2.2.2 Traffic Conflict Indicators Traffic conflict indicators are quantitative measures of the closeness of a conflicting pair of road users, in space and time, in anticipation to a point of collision (Ismail, 2010). A number of traffic conflict indicators can be found in the literature. Traffic conflict indicators were well documented in previous studies (Cunto, 2008; Ismail, 2010; Autey, 2012). The most common indicators are: Time to collision (TTC), Post-encroachment time (PET), Gap time (GT), and Deceleration to safety distance (DST). 2.2.2.1 Time to Collision (TTC) The initial definition of the time to collision (TTC) was proposed by Hayward (1972) as “the time required for two vehicles to collide if they continue at their present speeds and on the same path” (Hayward 1972). A later definition of TTC was presented by Amundsen and Hydén (1977) 14 as “an observable situation in which two or more road users approach each other in space and time to such an extent that there is a risk of collision if their movements remained unchanged”. TTC, or some versions based on TTC concept, may be considered as the most widely used conflict indicator. The TTC concept is based on predicting future positions of pair of road users to identify the time till the hypothesized occurrence of collision course between them. Different implementations of the TTC concept vary in the method of extrapolating positions of pair of users, and the instant of time at which the extrapolating applied. In the literature, an attempt to simplify TTC was referred as the time to accident (TTA) and defined by Hydén (1987) as “the time that passes from the moment that one of the road users reacted and starts braking or swerving until the moment the involved road user had reached the point of collision if both road users had continued with unchanged speed and direction”. In other words, instead of continuous measurements of TTC, only single value of TTC is recorded at the instant the evasive action took place (Cunto, 2008). The TTA calculation assumes vehicles continue at their current velocity. The Swedish TCT developed at Lund Institute of Technology derives a measure of severity by dividing the TTA by the closing speed which is the instantaneous velocity of the vehicle taking evasive action at the time this action is commenced (Svensson, 1998; Autey, 2012). For TTC estimation, previous studies (Saunier and Sayed, 2008; Saunier et al., 2010; Autey et al., 2012) used a probabilistic approach to define a vehicle’s future trajectory. In this approach, the probabilistic function is based on common motion patterns for prediction of a vehicle's future position. Computer vision techniques were used to produce common motion patterns and to assign a vehicle's trajectory probabilistically. A road user’s instantaneous velocity is used to extrapolate its position along the assigned motion paths (Autey, 2012). 15 Minderhoud and Bovy (2001) Presented two safety indicators which can be considered as extensions to the TTC indicator: 1) time-exposed-TTC (TET), and 2) time-integrated-TTC (TIT). The extended indicators can give a more complete and comprehensive picture of the safety level on a particular section of road during a particular period of time. The developed TET indicator expresses the exposition time to safety-critical approach situations, whereas the developed TIT indicator additionally takes into account the encountered TTC values during these safety-critical approaches. TET is a summation of all moments (over a considered time period) that a driver approaches a front vehicle with a TTC-value below a specific threshold value of TTC which is considered to be the boundary between safe and safety-critical approaches. Thus, the lower the TET value, the more safe the situation. TET does not take into account the variation in safety levels of different time-to-collision values below the threshold value; therefore, the TIT indicator has been developed. The TIT integrates the difference between the TTC value and the threshold value over all time periods that the TTC is below this threshold value. Thus, the TET and TIT combination accounts for both the exposure duration and the proximity of collision at all instants. Advantage of the TET and TIT safety measures above the conventional TTC measure is the inclusion of time-dependent time-to-collision values of all subjects that use a road section during a time period (Minderhoud and Bovy, 2001). To identify a conflict or to distinguish between near-misses and undisturbed passages, several researchers tried to establish minimum value of TTC to be used in estimating the number of conflicts in field studies (Brown, 1994; Hayward, 1972; van der Horst and Brown, 1989; van der Horst and Hogema, 1993). van der Horst (1990) concluded that the distinct detection threshold of TTC to discriminate between normal and critical encounters is 1.5 second. This 16 result was obtained from an experiment involving the application of a driver simulator to a closed course road (van der Horst, 1990; van der Horst and Hogema, 1993). Two main problems limit the application of TTC as a safety indicator or measure. First, several combinations of speed and distance can produce the same TTC measure which may not be of similar severity. Second, continuous measurements of TTC require detailed trajectories of vehicles to be presented in time and space which is difficult to achieve; however, this can be done through some techniques such as computer vision video analysis (Saunier and Sayed, 2007). 2.2.2.2 Post-encroachment Time (PET) Another common safety indicator is post encroachment time (PET). PET is defined as the time difference between the moment an “offending” vehicle leaves the area of potential collision and the moment the other vehicle arrives the collision area (Cooper, 1984). PET is less ambiguous than TTC as it requires no projection of road-user positions into the future. PET values can be measured discretely through the observation of vehicle trajectories (Autey, 2012). The main disadvantage of PET is that it does not require speed and distance measurement, hence missing many cues of conflict severity (Archer, 2004, Ismail, 2010). Another shortcoming of PET is the difficulty in identifying the willingness of the drivers to accept the risk (Chin and Quek, 1997). Another problem is that PET can be calculated even there is no possibility of a collision course. This usually happens in rear-end and merging conflict situations if the following vehicle has a lower or equal speed to the lead vehicle (Archer, 2004). 2.2.2.3 Gap Time (GT) Gap Time (GT) is given by the projected time arrival of the vehicle in the main traffic stream reaches the conflict area minus the time required for the yielding vehicle to clear the conflict 17 area. The projected time arrival of the main road vehicle is estimated using distance and speed relatives to the moment which the yielding vehicle begins the maneuver. This can also be described as the expected PET should road-users’ trajectories and velocities remain unchanged (Cunto, 2008; Autey, 2012). 2.2.2.4 Deceleration-to-safety Time (DST) Related with PET, Hupfer (1997) proposed the deceleration to safety time (DST) which is defined as the required deceleration for a vehicle to attain a non-negative gap-time in relation to another road-user. In other words, for a pair of road users with a calculable gap time, it is the deceleration that one vehicle must undertake to arrive immediately following the other road user (Autey, 2012). The above mentioned index has several applications. The primary application of this index is the vehicle-pedestrian interactions where a vehicle must decelerate to avoid hitting a pedestrian (Cafiso et al., 2010). 2.2.3 Challenges to Traffic Conflict Technique Despite the aforementioned advantages of the TCT, the TCT is commonly criticized for the consistency in conflict definition, the validity of the TCT, the high cost of data collection and the reliability of the manual collection of the conflict data. The main problems of the TCT have been well documented in (Chin and Quek, 1997) and can be summarized in the following subsections. 2.2.3.1 Consistency in Conflict Definition The first concept of traffic conflicts proposed by Perkins and Harris (1968) is based on the evasive actions taken by drivers. This can lead to many inconsistency problems in defining conflicts. First, some locations may have many categories of evasive actions associated with complicated conflicts; hence the conflict recording by field-observers would be difficult and prone to errors. Second, not all abrupt driver actions are necessarily evasive in reality. In fact, the 18 driver may apply the brakes for any other reasons and not as an evasive action to avoid a collision. Third, some logical problems are associated with equating evasive actions to conflicts especially when traffic conflicts are representing a surrogate measure to collisions. Many collisions and near misses happened because the driver had failed to take any evasive action. On the other hand, many recorded evasive actions are, in fact, precautionary ones. Thus, defining conflicts as evasive actions may lead to inconsistency in recognizing conflicts, and may cause a little correlation between conflicts and collisions. Traffic conflict definition by Amundsen and Hydén (1977) does not rely on evasive actions and provides some consistency to the conflict definition, however the distinction between conflict and non-conflict situation is still unclear in practice. To overcome the problem associated with the inconsistency in conflict definition, one way is to select only the serious cases of conflicts. To distinct between serious cases and non-serious cases, many previous studies (e.g. Guttinger, 1977; Hydén, 1977) have used time or distance thresholds. However, these thresholds appeared to be arbitrarily prescribed. Also, the TCT was proposed to remedy the problem of the rare collision data. This advantage may be lost if only the serious conflicts are selected for analysis. 2.2.3.2 Validity of TCT Validity of the TCT is often investigated by the statistical correlation between the number of observed conflicts and the number of recorded collisions. The validity became an issue of debate after a good number of studies failed to find an acceptable statistical correlation between conflicts and collisions. Some researchers such as (Glennon et al., 1977; Williams,1981) have claimed that for every case of good correlation, there was another of poor correlation. 19 Therefore, they questioned the usefulness of the technique and called for a reassessment of the whole concept of traffic conflicts. On the other hand, statistically significant correlations between conflicts and collisions have been found in previous studies (Brown, 1994; Sayed and Zein, 1999, El-Basyouny and Sayed, 2013). Also, a good number of researchers (e.g. Oppe, 1977; Hydén et al., 1982; Baguley, 1982; and Muhlrad, 1982) assumed the validity of the TCT, and sought to improve the correlations by redefining conflicts or explain the poor correlations between conflicts and accidents. Some researchers (Baker, 1972; Amundsen and Hydén, 1977) recommended that the TCT should be restricted to cases where collisions data are too low for traditional safety analysis. 2.2.3.3 Reliability in Conflict Measurements Traffic conflicts, in tradition, have been manually observed in the field by trained observers. There are two general aspects in such a method that lead to unreliability of the measurements. The first aspect comes from the inconsistency in recording made by individual observer (called intra-rater variation). The second one comes from the variability in interpretation and recording of conflicts between different observers (called inter-rater variation) (Chin and Quek, 1997). Inconsistency in conflict measurements by an individual observer can be caused by many reasons such as poor definition of conflict, excessive and complex conflicts, human factors (i.e. fatigue and lack of training). To remedy some of these problems, several countries developed manuals and training packages to detail the different types of conflicts and the required observation procedure (Glauz and Migletz, 1980; Hummer, 1994). As well, several studies tried to unify the different methods of conflict observation in order to ensure that the conflict experiment is repeatable (e.g. Malaterre and Muhlrad, 1979; Grayson et al., 1984). However, the subjectivity problem still exists in the manual field observations. 20 Video recordings can facilitate repeated viewing, and this may avoid some of the reliability problems. However, the video recording is limited by the two image dimensions and does not provide the same quality of observation as the observer on the field. Therefore, the field-observer has better judgment of conflict observation (Chin and Quek, 1997). A more promising alternative to the subjective and rule-based definitions of traffic conflict is objective conflict indicators such as TTC. With advances in the field of computer vision, these conflict indicators can be calculated automatically, hence increasing accuracy and reducing the high cost of data collection (Ismail, 2010). 2.3 Automated Video-based Computer Vision Techniques Recently, video-based automated computer vision techniques were shown to be useful for automated traffic conflict detection and in conducting various safety analysis applications such as before and after safety evaluations and studying road user behavior (Autey et al., 2012; Sayed et al, 2012; Sayed et al., 2013). The video-based computer vision approach can have many advantages. First, automated video-based analysis overcomes the shortcomings associated with the manual collection of conflict data in terms of the cost and the data reliability. Second, automated analysis of field conflicts depends on tracking vehicles trajectories; and therefore it can have more accuracy in terms of determining conflict severity (e.g. TTC) and conflict location. Third, video cameras have the ability of collecting rich and detailed traffic data. Fourth, the installation of video cameras is easier than the installation of the other sensors such as the magnetic loop detectors which requires the road surface to be excavated (Saunier and Sayed, 2006). 21 2.3.1 Tracking of Road Users Many researchers have been developing algorithms or systems for tracking road users from video recordings using advanced methods (Gupte et al., 2002; Laureshyn et al. 2009; Malinovskiy et al. 2007; Atev et al., 2005). In the literature, there are four main approaches for object tracking: 1) 3D model-based tracking; 2) Region-based tracking; 3) Contour-based tracking; and 4) Feature-based tracking (Saunier and Sayed, 2006; Cavallaro et al., 2005). 2.3.1.1 3D Model-based Tracking Model-based tracking uses the prior knowledge of typical objects in a given traffic scene especially with parametric three dimensional object models. These methods recognize vehicles by matching a projected model to the image data. Then the trajectories and models can be recovered with high accuracy for a small number of vehicles. In doing so, the problem of partial occlusion can be addressed (Lou et al., 2005). The main shortcoming of this method is the reliance on detailed geometric object models. This means that detailed models for all vehicles, in the field of study, should be accurately built before using the algorithm (Malinovskiy et al., 2007; Saunier and Sayed, 2006). 2.3.1.2 Region-based Tracking Region-based tracking depends on identifying connected regions of the image (blobs) associated with each road user. This can be obtained through background subtraction, and then tracked over time using region information (such as motion, size, color, shape, texture, and centroid). Although Region-based tracking approach works well in free-flowing traffic, it can lead to multiple vehicles being grouped together as one large blob in congested traffic (Saunier and Sayed, 2006). 22 2.3.1.3 Contour-based Tracking In contour-based tracking, the contour of a moving object is represented by a continuous line which is updated dynamically relying on the boundary curves of the moving object. This method provides more efficient description of objects and less computation complexity than region-based method. However, the problem of occlusion remains (Saunier and Sayed, 2006; Autey, 2012). 2.3.1.4 Feature-based Tracking Instead of tracking objects as a whole, feature-based tracking tracks distinguishable points or lines on the object (features). This may partly overcome the problem of occlusion. Also, the same algorithm can be used for tracking in daylight, or night-time conditions, as well as different traffic conditions. Features are identified and tracked using an implementation of well-developed methods such as Kalman filtering and the well-known Kanade-Lucas-Tomasi Feature Tracker algorithm. In feature grouping, the main purpose is to decide what set of features belongs to the same object. Features that move at similar speed and satisfy other spatial proximity and common motion constraints are grouped to create objects where the trajectories of these objects are recorded (Saunier and Sayed, 2006). This tracking approach is used for the computer vision automated traffic analysis system developed in the Transportation Group at UBC. 2.3.2 Computer Vision at UBC There is continuous research in the Transportation Engineering Group at the Civil Engineering Department of UBC that aims to develop an automated road safety analysis system based on video sensors. This video analysis system is based on existing state-of-the-art computer vision algorithms and has incorporated some adaptations for the study of road users (Autey, 2012; Ismail, 2010). 23 A detailed description of the automated computer vision video analysis process, validation and its past applications in safety assessment is presented in (Autey et al., 2012; Saunier and Sayed, 2006; Saunier and Sayed, 2007; Saunier and Sayed, 2008; Saunier et al., 2010; Zaki et al., 2013). Camera calibration is applied to identify a mapping between the three-dimensional real world and the two-dimensional image space. Camera calibration is necessary for relating road user tracks to real-world positions (Ismail et al., 2013). The calibration process can be performed based on feature correspondences between the video image and the orthographic image of the intersection. The road-user detection and tracking module relies on a feature-based tracking method described in (Saurnier and Sayed, 2006). Feature tracking involves tracking distinguishable points (features) on moving objects. The automated video analysis relies on computer algorithms to differentiate between features of road users and features that are part of the environment described in (Saunier and Sayed, 2006). Features are identified and tracked using an implementation of the well-known Kanade-Lucas-Tomasi Feature Tracker algorithm. In feature grouping, the main purpose is to decide what set of features belongs to the same object. Features that move at similar speed and satisfy other spatial proximity and common motion constraints are grouped to create objects where the trajectories of these objects are recorded. The tracking accuracy for this approach has been tested and found to be about 90% which is considered reliable (Ismail et al., 2010) and should have little impact on the accuracy of the estimation of conflict indicators. Prototypes refer to a group of motion patterns that define the set of movements carried out by road users in the video scene. The road user movements are matched to those prototypes and trajectories that show potential conflicts are identified. The trajectory of an object is matched 24 to individual prototypes from the full set of prototypes using a Longest Common Sub Sequence (LCSS) algorithm. This provides a set of predicted future positions with associated probabilities of occurrence. Conflicts between vehicles can then be determined by evaluating if any of these future positions coincide spatially and temporally with other vehicles. The conflict analysis involves the calculation of conflict indicators such as the Time-to-Collision Conflict (TTC). The minimum TTC can be extracted to represent the conflict severity. A critical value of a conflict indicator must be drawn from each interaction. Typically, the most severe value is used to represent the overall severity of a traffic event. 2.4 Traffic Microsimulation Models and SSAM The use of traffic simulation models for conducting conflict-based safety evaluations has also been recently proposed (Gettman et al., 2003). There is a growing interest in using simulation models for the safety assessment of road facilities by analyzing simulated vehicle trajectories and estimating conflict indicators. The main advantages of this approach are: 1) the ability to evaluate the safety of various design and traffic management options of road facilities before actually making changes and 2) the ease of estimating simulated conflicts without actually observing them. 2.4.1 Surrogate Safety Assessment Model (SSAM) The Surrogate Safety Assessment Model (SSAM) was recently proposed by (Siemens Energy & Automation, Inc.) and was sponsored by the Federal Highway Administration (FHWA) of the United States. The SSAM was developed to estimate traffic conflicts using simulated vehicle trajectories exported from four commonly-used microscopic simulation models: VISSIM, AIMSUN, PARAMICS, and TEXAS. Several traffic conflict indicators can be calculated including the time to collision (TTC), post-encroachment time (PET), deceleration 25 rate (DR), maximum speed (MaxS), and speed differential (DeltaS). In SSAM, conflicts are classified into three maneuver types: rear-end; lane-change; and crossing. Conflicts are identified based on conflict angle and specific thresholds for TTC and PET, which are predetermined by the user (Gettman et al., 2008). 2.4.2 Calibration of Microsimulation Models Several previous studies have examined the calibration and validation of microsimulation models for use in traffic safety evaluations. Gettman et al. (2008) conducted theoretical and field validation to assess the use of SSAM. The theoretical validation was conducted to evaluate the relative safety of pairs of design alternatives in 11 case studies that include intersections and interchanges. The results found that SSAM could recognize statistically significant differences in the total number of conflicts, the number of conflicts by type, and conflict severity indicators for both design alternatives under equivalent traffic conditions. The field validation was also conducted to evaluate the use of SSAM as a surrogate safety assessment tool. For the field validation, 83 signalized intersections, modeled in VISSIM and assessed with SSAM, were used. The simulated conflicts of these intersections were compared to the actual crash data using 5 statistical tests. The results found a statistically significant correlation between the simulation conflicts data provided by SSAM and the field crash data. However, it should be noted that this correlation maybe expected due to both simulated conflicts and crashes being correlated with traffic volume. As well, a significant correlation between the ranking of intersections with abnormally high conflicts and abnormally high crashes was not found. Therefore, the validation results were considered inconclusive, and further research was recommended to improve driver behavior in simulation models, develop a composite “safety index”, study the nature of conflicts in real-world data, and collect adequate vehicle trajectory data sets (Gettman et al., 2008). 26 Dijkstra et al. (2010) conducted a study at 569 junctions in the Netherlands. The microscopic simulation model PARAMICS was used to model the traffic of the study area. The relationship between observed crashes and simulated conflicts was assessed by using generalized linear models. Different log-linear models were developed by assuming either a negative binomial or a Poisson distribution of the crashes. Various TTC thresholds were used in the calculation of conflicts. Regression analyses were performed, and the two different distributions (Poisson and negative binomial) led to similar results in terms of the goodness of fit of the models and the significance of the variables. It was found that the number of conflicts at junctions and the number of passing motor vehicles are statistically related to the number of observed crashes (Dijkstra et al., 2010). Cunto and Saccomanno (2008) presented a systematic procedure for calibrating a microscopic simulation model at signalized intersections based on a safety performance measure (crash potential index), that relates deceleration rate required to avoid a crash (DRAC) to the maximum available deceleration rate (MADR) of vehicles. The microscopic traffic algorithms were obtained from VISSIM to simulate safety performance, and the real trajectories of vehicles extracted from Next Generation Simulation Program (NGSIM) administered by the Federal Highway Administration (FHWA). The calibration process included four steps: 1) heuristic selection of initial model inputs; 2) statistical screening using factorial design; 3) development of a linear expression relating significant model inputs to safety performance (fractional factorial analysis); and 4) genetic algorithm application to obtain best estimates of model parameters. The model results were validated, and their transferability was investigated, using an independent sample of vehicle tracking data (Cunto and Saccomanno, 2008). 27 Duong et al. (2010) proposed a multi-criteria calibration procedure for parameters calibration of microscopic traffic simulation platforms. The multi-criteria procedure considered the underlying traffic attributes (i.e. volume, speed, and density), and provided reasonable estimates of safety performance for the real trajectories dataset from NGSIM (FHWA) (Duong et al., 2010). Sobhani et al. (2013) used VISSIM to develop a simulation based modelling approach to assess the safety performance of road locations. The developed framework consists of two main components. The first component was to estimate the number and the severity of conflicts using micro simulation models. The second component was to measure the potential injury severity of each simulated conflict (Sobhani et al., 2013). Fan et al. (2013) used VISSIM and SSAM for estimating field-measured traffic conflicts at freeway merge areas. Field conflicts were collected manually and compared with simulated conflicts. A two-step calibration procedure was proposed to calibrate and validate the VISSIM simulation models. In the second stage of calibration, the minimum TTC threshold in SSAM was calibrated to give minimum mean absolute percent error (MAPE) of simulated rear-end conflicts. The optimum TTC threshold was found to be (2.1 seconds). The transferability of the calibrated simulation models was then tested using another dataset, which was not used in the calibration process. It was found that the two-stage calibration procedure reduced the MAPE for rear-end and lane-change conflicts. The results showed that there was a reasonable consistency between the simulated and the observed conflicts (Fan et al., 2013). Huang et al. (2013) compared the simulated conflicts generated by the VISSIM and SSAM to manually collected field conflicts at 10 signalized intersections. A two-stage calibration procedure was proposed to improve the consistency between the simulated and the 28 observed conflicts. In the second stage of calibration, the minimum TTC threshold in SSAM was calibrated to give minimum MAPE of simulated rear-end conflicts; and the minimum gap time in VISSIM was calibrated to give minimum MAPE of simulated crossing conflicts. The optimum TTC threshold, and the optimum minimum gap time were found to be (1.6 seconds), and (2 seconds), respectively. The transferability of the simulation model was tested by applying the calibrated parameters in another two sites without recalibration (Huang et al., 2013). In the two aforementioned studies, the field-measured conflicts were collected manually which may lead to accuracy issues. In addition, the SSAM TTC threshold which is used to identify rear-end conflicts, was used as a calibration parameter. Therefore, the calibration process was about finding a TTC threshold that will minimize the MAPE between the number of the simulated and field-measured conflicts. So et al. (2015) used an integrated simulation environment to assess the safety impact of the connected vehicles- based driver warning systems. This integrated simulation approach consists of two parts: real time simulation approach and post-processing approach. In the real-time simulation environment, a driver warning simulator, the GPS simulator, and communication delays simulators were developed and then integrated with VISSIM. The vehicle trajectories produced from this real-time simulation were post-processed by the vehicle dynamics model to generate the vehicle dynamics-based trajectories. Finally, SSAM estimated traffic conflicts using the vehicle dynamics-based trajectories (So et al., 2015). Stevanovic et al. (2015) presented a new method for signal timing optimization through a tradeoff between mobility, safety, and the environment. The method integrated the traffic simulation model (VISSIM); surrogate safety assessment (SSAM); fuel consumption and emission modeling; and signal timing optimization tool (Stevanovic et. al., 2015). 29 Chapter 3: Field-Measured Data This chapter contains two sections. The first section describes the study location and the field data collected for analysis. The second section provides information on the automated video-based computer vision analysis used to identify field traffic conflicts. 3.1 Study Location and Data Collection Two urban signalized intersections were selected for this study. The both intersections are located on the same corridor: 72nd avenue in the city of Surrey, British Columbia, Canada. Figure 3.1 shows the intersections location. Both of them are a signalized intersection with 4 protected-permissive left turns. The data from the first intersection was used to perform the proposed calibration procedure of the simulation model; while the data from the second intersection was used to investigate the transferability of the calibrated parameters of the simulation model. Figure 3.1 Study location 3.1.1 The First Intersection The intersection of 72nd avenue and 128th street in the city of Surrey, British Columbia, Canada is the first intersection selected for this study. Traffic videos were recorded using 8 cameras distributed to cover 8 sections over the 4 approaches of the intersection (2 sections per each 30 approach). Figure 3.2 illustrates the intersection location and cameras’ positions. A computer program was developed to extract the actual signal program for each hour by identifying the different colors of the traffic signal from video frames. Also, a detailed count of traffic volumes for each hour was extracted from the recorded videos and included all movements in all directions, number of vehicles arriving during the green time, traffic composition, and number of public transit buses. Using vehicle trajectories, another computer program was developed to estimate travel time for each approach each hour. In addition, the actual desired speed cumulative distribution curve was determined. The desired speed is the speed chosen by a vehicle if not hindered by other vehicles or traffic control devices. Therefore, to get the desired speed distribution from field data; speeds of 1086 vehicles, selected during green time and low-volume periods, were calculated using vehicle trajectories. Figure 3.3 shows the cumulative distribution curve of the desired speed. Finally, the average delay time per hour for each approach was calculated by comparing the travel time for each vehicle with its ideal travel time. The ideal travel time was based on the travelled distance by the vehicle and the desired speed (using desired speed distribution, average desired speed was assumed). Traffic data details are provided in Appendix A. 31 Figure 3.2 Cameras positions at the first intersection At this intersection, a total of 72 hours of video recordings (9 hours X 4 approaches X 2 sections) were recorded in two consecutive days in March 2012. In delay estimation, 4 approaches representing 36 hours (9 hours X 8 sections /2 = 36) were used (Each couple of sections represents one approach), and classified into 18 hours for calibration (East and west approaches); and 18 hours for validation (North and south approaches). In conflicts calibration, 60 hours were selected from 72 hours and classified into: 30 hours for calibration (East and west approaches (4 sections)); and 30 hours for validation (North and south approaches (4 sections)) (see Chapters 3 and 4). 32 Figure 3.3 Cumulative distribution curve of desired speed 3.1.2 The Second Intersection The intersection of 72nd avenue and 132nd street in the city of Surrey, British Columbia, Canada is the second intersection selected for this study. Traffic videos were recorded using 4 cameras distributed to cover 4 sections over the 2 approaches located on the 72nd avenue corridor (2 sections per each approach). Figure 3.4 illustrates the intersection location and cameras’ positions. All traffic data including: traffic signal program; traffic volumes; number of vehicles arriving during the green time; traffic composition; number of public transit buses; desired speed; and average delay were extracted from videos as the first intersection. At this intersection, a total of 36 hours of video recordings (9 hours X 2 approaches X 2 sections) were recorded in two consecutive days in March 2012. In transferability analysis, the 36 hours were filtered to 23 hours and used in investigating the transferability of the calibrated parameters of the simulation model (see Chapter 6). Traffic data details are provided in Appendix A. 33 Figure 3.4 Cameras positions at the second intersection 3.2 Automated Video-Based Computer Vision Analysis This section provides details on the automated analysis of video data using the automated video-based computer vision analysis tool developed at UBC. A detailed description of the automated video analysis process, validation and its past applications in safety assessment is presented in (Autey et al., 2012; Saunier and Sayed, 2006; Saunier and Sayed, 2007; Saunier and Sayed, 2008; Saunier et al., 2010; Zaki et al., 2013). The video analysis procedure is shown in Figure 3.5. A detailed description of the video analysis and safety diagnoses for the two intersections selected for this thesis is provided in (Tageldin et al., 2014). 34 Figure 3.5 A schematic diagram of the process of video analysis The following subsections show in detail the followed steps of the video analysis procedure applied for both selected intersections. 3.2.1 Video Encoding The first step in the video analysis was to do the video conversion of the recorded raw video footage to a format which can be used in the analysis. The encoding was performed to define the video scenes, hours, and days recorded that helps in assisting the analysis of each day. 35 3.2.2 Camera Calibration During video recording, the three-dimensional real-world is captured on a two-dimensional image space. This translation of three-dimensional coordinates into two-dimensional coordinates is a linear transformation that is associated with the properties of the camera and its lens. The linear transformation is defined by a matrix, called the homography matrix, which is related to the camera’s extrinsic parameters (camera position and orientation), as well as its intrinsic parameters (focal length, skew angle, and radial lens distortion). Camera calibration is the process of determining the homography matrix of a camera angle, and is necessary for tracking vehicles in the camera image and relating these tracks to positions in the real world. Details of the camera calibration approach are presented in previous work (Ismail et al., 2013). The calibration process began with the annotating of features in the camera image and in an aerial, orthographic image of the intersection. Google Satellite images were used in this study as shown in Figure 3.2 and Figure 3.4. Four types of annotations were used for camera calibration optimization as shown in Figure 3.6. 36 Figure 3.6 Four types of annotations used for camera calibration process (Source: Tageldin et al., 2014) 3.2.2.1 Corresponding Points Distinguishable points are identified and annotated in both the camera image and orthographic image. The selected points must be on the plane of the road or at a specified height above the road plane. The optimization algorithm adjusts the camera calibration so that when points are projected from the camera image to the orthographic image, or vice versa, the corresponding points match. 3.2.2.2 Distances The known real-world distance between two points in the camera image, both situated on the road surface plane, can be annotated during calibration process. The real distances were 37 measured from Google satellite images. The optimization procedure adjusts the camera calibration so that the projections of the end points into the orthographic view are the correct distance apart. 3.2.2.3 Angles Known angles between parallel lines of lane markings or the right angle lines at the intersections were used for camera calibration. The optimization procedure adjusts the camera calibration so that the projection of these lines into the orthographic view achieves the specified angle. 3.2.2.4 Global Up Directions The many poles mounted near intersections can provide information for retrieving three-dimensional coordinates from an image. It is assumed that poles are correctly mounted and are perpendicular to the road surface. By specifying the locations in the camera image of the tops and bottoms of poles, the calibration optimization algorithm can use this information to help determine the tilt of the road surface plane. 3.2.3 Calibration Verification The calibration error was represented by the discrepancy between calculated and annotated segment lengths normalized by the length of each segment. The accuracy of the final estimates was very good and no further error in conflict analysis was attributed to inaccurately estimated camera parameters. Hence, the camera calibration step was done, the real-world position of points in the image can be recovered, and video analysis can be undertaken. 3.2.4 Feature Tracking The automated video analysis relies on computer algorithms to differentiate between features of road users and features that are part of the environment. Features are identified and tracked using 38 an implementation of the well-known Kanade-Lucas-Tomasi Feature Tracker algorithm. Tracked features are further refined through the addition of filters. First, features that remain stationary are assumed to belong to the environment and are discarded and not tracked (Saunier and Sayed, 2006). For this reason, features must be continually generated in order to identify moving features that may subsequently be tracked. Second, feature-tracker errors are dealt with by enforcing regularity motion checks to remove features showing unreasonable acceleration or sudden changes in direction. These features are displaying motion properties that are not physically possible of road users and can safely be classified as tracking errors. 3.2.5 Feature Grouping (Objects Creation) Vehicles are large objects with many distinguishable physical features and, as such, will generate multiple features during the feature-tracking procedure. The next step is to decide which set of features belongs to a unique vehicle so an object may be generated. Feature grouping is carried out using cues like spatial proximity and common motion of features. Among a detailed set of criteria, two important tests are the connection distance between features and the similarity of the motion-vectors features. For a feature to be added to another to create, or add to, a group, it must be within the maximum connection distance pre-specified by the user. In the real world, features of a vehicle have identical motion vectors due to the physical rigidity of the vehicle. Computer-tracked features must exhibit the same characteristic in order to be associated with a common vehicle. Features with motion vectors differing by more than a specified threshold are assumed to not belong to the same vehicle and are not grouped, regardless of their spatial proximity. A detailed description of the tracking and grouping algorithm is presented in (Saunier and Sayed, 2006). The tracking accuracy for motor vehicles has been measured between 84.7% and 94.4% on three different sets of sequences. This accuracy is considered reliable especially under heavy 39 traffic flow conditions and should have little impact on the accuracy of the calculation of conflict indicators. This means that most trajectories are detected by the system and the calculated conflict indicators are considered reliable. 3.2.6 Prototype Generation and Matching Prototypes refer to a group of motion patterns that define the set of movements carried out by road users in the video scene. The road user movements are matched to those prototypes and trajectories that show potential conflicts are identified. The trajectory of an object is matched to individual prototypes from the full set of prototypes using a Longest Common Sub Sequence (LCSS) algorithm with a maximum LCSS matching distance. An object will therefore be matched with more than one prototype with a probability weighting determined from the LCSS matching distance. The matched prototypes are translated to the object’s center and corrected for the current velocity of the object. This provides a set of predicted future positions with associated probabilities of occurrence. 3.2.7 Conflicts Analysis Conflicts between vehicles can then be determined by evaluating if any of these future positions coincide spatially and temporally with other vehicles (Saunier et al., 2010). Every pair of road users that share temporal and spatial proximity has the potential for a conflict. Two vehicles that appear at the same time, both inside the field of view of the camera are defined as having spatial and temporal proximity, and potential conflicts between them calculated. A critical value of a conflict indicator must be drawn from each interaction. Typically, the most severe value is used to represent the overall severity of a traffic event. The Time to Collision (TTC) conflict indicator is commonly implemented as a measure of the severity of a conflict. TTC is continually calculated between conflicting vehicles, and thus 40 a set of values is returned for each discrete conflict. One representative value, the minimum TTC, is extracted from this set to indicate the maximum severity of this interaction. The minimum TTC from all conflicting vehicles and the total number of events with minimum TTC less than three seconds were recorded. Due to potential noise in road user tracks, different filtering strategies have been tested. Furthermore, the first and last 10 frames of each track were discarded and selected events were visually reviewed to identify any tracking errors. In this study, only traffic events associated with minimum TTC of less than 3 seconds were considered. This value was selected based on the work of (Sayed and Zein, 1999). Validation was performed on a subset of the events selected from the interactions database. The scope of the validation was limited to a comparison between an event’s minimum TTC and a corresponding manually calculated TTC. The results demonstrated the accuracy of the automated TTC index estimation. By knowing the minimum TTC of each conflict, the number of real rear-end conflicts was determined at 21 thresholds of TTC for 60 section-hours for the first intersection, and 23 section-hours for the second intersection. Thresholds of TTC ranged from 1 second to 3 seconds in 0.1 second increments. Also, conflict locations determined using the coordinates (x, y) of the critical event position (minimum TTC). Details of the field-measured rear-end conflicts, for both selected intersections, are provided in Appendix B. 41 Chapter 4: The Microsimulation Model VISSIM This chapter contains six sections. The first section provides general information on the proposed calibration framework of the simulation model (VISSIM). The second section explains the establishment of the analyzed intersection using VISSIM platform. The third section includes all details of the proposed two-step calibration procedure of VISSIM and the calibration results. The fourth section includes the details and the results of the validation of the proposed procedure. The fifth section includes the spatial distribution of the conflicts using heat maps. The sixth section provides a summary and discussions for this chapter. 4.1 VISSIM Calibration Framework The data from the first intersection was used to develop, calibrate, and validate a microsimulation model. The well-known microsimulation model VISSIM and SSAM tool were used to estimate the simulated conflicts. The VISSIM model of the first intersection was developed, and then a two-step calibration procedure was carried out. The calibration procedure was validated using another dataset (other approaches at the first intersection) had not been used in the calibration. Figure 4.1 presents the general framework of the calibration and validation processes. Finally, conflicts heat maps were provided to compare simulated and field-measured conflicts locations. 42 Figure 4.1 General framework of the simulation model calibration process 4.2 VISSIM Platform VISSIM is a widely used microscopic simulation model. VISSIM is a time-step and behavior-based model developed to simulate traffic and depends on a psycho-physical car-following model based on the Wiedemann models (Wiedemann, 1974; Wiedemann, 1991) which assume that the driver can have one of four driving modes: free driving, approaching, following, and braking (VISSIM 5.30 User Manual, 2011). In this study, in order to assess the intersection safety using simulation, the VISSIM model has to be developed accurately to match actual field conditions. First, the geometry of the intersection (e.g. number of lanes, lanes widths, and turning radius) was extracted from an aerial photo and then drawn in a technical drawing. Once the geometry technical drawing was defined in the VISSIM model, all links and connectors were set up with the real dimensions. For each hour, the detailed traffic count extracted from videos was defined in VISSIM using routes to 43 represent all movements (i.e. through, left, right), and traffic composition (i.e. percentage of passenger cars, trucks, and buses). Public transit was defined in the VISSIM model using the real number of public transit buses extracted from videos. Actual traffic signal settings were defined in VISSIM for each hour using a Ring Barrier Controller (RBC). In RBC, in addition to the signal cycle length, each phase has minimum green time, maximum green time, yellow time, and all red time. To represent the protected-permissive left turns, a detector was defined at each one of the 4 left-turn lanes. In each signal cycle, if the detector is occupied, a protected left turn phase will be called by RBC, otherwise the left turn will be permissive. Figure 4.2 illustrates the developed VISSIM model for the first intersection. Finally, visual inspection was done to check the simulation and to ensure that there is no unrealistic conditions. Trajectory files were exported from VISSIM and processed in SSAM to estimate the simulated rear-end conflicts at the 21 thresholds of TTC for the same 60 section-hours (for the first intersection) pre-estimated using the automated video analysis. The results of average delay time and rear-end conflicts will be discussed later in this chapter. 44 Figure 4.2 VISSIM model for the first intersection 4.3 The VISSIM Calibration Procedure As mentioned earlier, simulation models such as VISSIM need to be well calibrated to give reasonable and realistic results. In this study, a simplified two-step calibration procedure of VISSIM driving behavior parameters was proposed to enhance the correlation between the simulated and the field observed rear-end conflicts. The first intersection was used for the calibration process. The east and west approaches (i.e. 72nd avenue) were used as a calibration dataset. 18 hours (9 hours X 2 approaches) were used to calculate average delay time and 30 45 section-hours (i.e. sections 1, 3, 5, 7 in Figure 3.2) were used to calculate rear-end conflicts. In the first calibration step, the VISSIM model was calibrated to ensure that the simulation gives reasonable results of average delay times. Then, in the second calibration step, VISSIM parameters were calibrated to enhance the correlation between the simulated and the real rear-end conflicts. 4.3.1 First Step Calibration The main goal of this step is to match the simulated average delay times with the field-observed average delay times. Other measures such as the number of stops, queue length, or a combination of these measures can be used as a target of this calibration step. However, optimizing several measures at the same time can be difficult. Therefore, the average delay per vehicle was selected for optimization as it is the primary measure in evaluating the LOS for signalized intersections (Highway Capacity Manual, 2000). In order to calibrate the average delay time, the arrival type and desired speed were calibrated to match the field conditions. The arrival type for each lane group describes the quality of the progression which is a critical characteristic that must be quantified for the analysis of an urban street or signalized intersection. In order to quantify the arrival type, the platoon ratio needs to be determined .The platoon ratio depends on the proportion of all vehicles arriving during green and the ratio of effective green time for the lane group to the cycle length (Highway Capacity Manual, 2000). Based on the real signal time analysis and field proportion of vehicles arriving during green which were extracted from the video recordings, the platoon ratio and the arrival type were determined for each hour as an average. The VISSIM model was calibrated to produce similar arrival type for each approach for each hour. To match the field arrival type using the simulation model, a dummy signal head was added before the four approaches of the analyzed intersection. 46 Each dummy signal has offset time (Δt) from the simulation time. By changing the offset time, the percentage of vehicles (P) arriving on green time of the analyzed intersection can be changed. We can obtain the offset time which gives a value for P that matches the field condition. As well, we added a bypass at each approach to represent the percentage of vehicles (X %) arriving out of platoon. The percentage X was measured from the field data as average for each hour. The percentage X was defined in VISSIM by defining two routes: the bypass route which has X% of the vehicles; and the dummy signal head route which has (1-X) % of vehicles. Figure 4.3 illustrates the process of arrival type calibration. Figure 4.3 Arrival type calibration process (VISSIM) In VISSIM, the desired speed is represented by a cumulative distribution curve to account for its stochastic nature. In the first calibration step, the cumulative distribution curve for the simulation desired speed was modified in VISSIM to match the one extracted from videos. Figure 4.4 shows the different arrival types for 18 hours (9 hours X 2 approaches). It can be noted that the field platoon ratio varies from 0.53 to 1.46; however, the default VISSIM model gives the platoon ratio varying from 1.06 to 1.30. Figure 4.5 illustrates the average delay time for the 18 hours. After the first calibration, the mean absolute percent error (MAPE) value 47 for the average delay time was decreased from 31.7% to 11.5%. The MAPE value was determined by the following equation: 𝑀𝐴𝑃𝐸 =1𝑛 ∑𝑑𝑠𝑖𝑚 − 𝑑𝑟𝑒𝑎𝑙𝑑𝑟𝑒𝑎𝑙𝑛𝑖=1 In the previous equation, 𝑛 represents the number of hours, 𝑑𝑠𝑖𝑚 represents the average delay time in seconds for one hour from simulation, and 𝑑𝑟𝑒𝑎𝑙 represents the real average delay time in seconds for one hour. Figure 4.4 Arrival types for calibration dataset (VISSIM) 48 Figure 4.5 Average delays for calibration dataset (VISSIM) After the first-step calibration, the trajectory files were exported from VISSIM and imported in SSAM in order to estimate the simulated rear-end conflicts. Simulated rear-end conflicts for the selected 30 section-hours were estimated at the 21 thresholds of TTC. The correlation between the real and simulated conflicts at all TTC thresholds was determined and the results will be discussed later in section 4.3.3. 4.3.2 Second Step Calibration The main purpose of this step is to enhance the correlation between field-observed and simulated conflicts by calibrating VISSIM parameters. First, sensitivity analysis was done to determine important VISSIM parameters which have the biggest effect on the simulated rear-end conflicts. Subsequently, a Genetic Algorithm technique was applied to estimate the best values of these parameters. 49 4.3.2.1 Sensitivity Analysis VISSIM contains a large number of parameters (about 190) (VISSIM 5.30-05 User Manual, 2011) which makes the calibration process very complicated. However, the focus of this study is on driving behavior parameters as they control the way the simulated vehicles progress through the location. To simplify the calibration process, sensitivity analysis was conducted to determine the important parameters. Thirty driving-behavior parameters representing car-following model parameters, lane-change parameters, and signalized intersection parameters were considered in the sensitivity analysis. In addition, an additional parameter representing desired deceleration was considered based on results from a previous study (Cunto and Saccomanno, 2008). For each parameter, a range of values, which includes the default, was determined based on previous studies and engineering judgment. 357 simulation runs ((29 parameters X 4 values + 2 parameters X 1 value + 1 default) X 3 random seeds) were conducted and the simulated rear-end conflicts were estimated at each time at different thresholds of TTC. An Analysis of Variance (ANOVA) was applied to perform the sensitivity analysis. Table 4.1 summarizes the sensitivity analysis results and shows the most important 6 parameters. These include CC0 (i.e. standstill distance), CC1 (i.e. headway time), CC4 & CC5 (i.e. negative and positive following thresholds), reduction factor for safety distance closed to stop line, start upstream of stop line, and desired deceleration. Table 4.2 provides more details of these parameters in the context of the driving behavior in VISSIM. The six parameters affect the desired safety distance between each couple of consecutive vehicles in VISSIM, and can be considered as the safety-related parameters at signalized intersections. More description of the selected parameters can be found in (VISSIM 5.30 User Manual, 2011). Detailed sensitivity analysis calculations were provided in Appendix C. 50 ANOVA Test (Alpha = 0.01) Parameter Sum of squares Degree of freedom Mean square F-value P-value F critical Is significant? Following Parameters 1 CC0 (Standstill distance) 34819.33 4 8704.83 11.03 0.0000 3.53 YES 2 CC1 (Headway time) 24444.75 4 6111.19 3.83 0.0064 3.53 YES 3 CC2 (Following variation) 995.22 4 248.80 1.76 0.1447 3.53 NO 4 CC3 (Threshold of entering following) 2043.70 4 510.92 2.89 0.0263 3.53 NO 5 CC4&CC5 (Negative and positive following thresholds) 5081.38 4 1270.35 4.43 0.0026 3.53 YES 6 CC6 (Speed dependency of oscillation) 1047.80 4 261.95 2.26 0.0691 3.53 NO 7 CC7 (Oscillation acceleration) 995.08 4 248.77 1.27 0.2862 3.53 NO 8 CC8 (Standstill acceleration) 333.55 4 83.39 0.77 0.5488 3.53 NO 9 CC9 (Acceleration at 80 km/h) 1561.05 4 390.26 2.02 0.0984 3.53 NO 10 Look ahead distance (Min.) 8.53 4 2.13 1.52 0.2027 3.53 NO 11 Look ahead distance (Max.) 304.13 4 76.03 0.36 0.8356 3.53 NO 12 Observed vehicles 1008.72 4 252.18 1.98 0.1039 3.53 NO 13 Look back distance (Min.) 0.45 4 0.11 1.86 0.1233 3.53 NO 14 Look back distance (Max.) 546.62 4 136.65 1.92 0.1132 3.53 NO 15 Temporary lack of attention (Duration & Percentage) 288.47 4 72.12 0.54 0.7090 3.53 NO Lane Change Parameters 16 Max. deceleration (Own) 47.72 4 11.93 0.35 0.8468 3.53 NO 17 Max. deceleration (Trailing) 672.55 4 168.14 3.35 0.0132 3.53 NO 18 Own (-1m/s² per distance) 106.97 4 26.74 0.99 0.4186 3.53 NO 19 Trailing (-1m/s² per distance) 3.22 4 0.80 0.13 0.9719 3.53 NO 20 Accepted deceleration (Own)* * * * * * * NO 21 Accepted deceleration (Trailing)* * * * * * * NO 22 Waiting time before diffusion 15.80 4 3.95 0.30 0.8745 3.53 NO 23 Min. headway (front/rear) 293.38 4 73.35 1.00 0.4114 3.53 NO 24 Safety distance reduction factor 1047.80 4 261.95 1.46 0.2193 3.53 NO 25 Max. deceleration for cooperative braking 629.72 4 157.43 1.17 0.3272 3.53 NO Signal Control Parameters 26 Decision model 320.33 1 320.33 2.11 0.1602 7.88 NO 27 Behavior at red/amber signal* * * * * * * NO 28 Reduction factor 112008.37 4 28002.09 6.79 0.0001 3.53 YES 29 Start upstream of stop line 13765.95 4 3441.49 6.89 0.0001 3.53 YES 30 End downstream of stop line 701.80 4 175.45 1.41 0.2379 3.53 NO Functions 31 Desired deceleration 9206.80 4 2301.70 5.36 0.0006 3.53 YES * No change in rear- end conflicts Table 4.1 Sensitivity analysis results 51 Parameter Description Unit Default CC0 Standstill distance: the desired distance between stopped vehicles. This distance does not have variation through all the simulated vehicles. m 1.50 CC1 Headway time: the time that a driver wants to keep. The higher the value, the more cautious the driver is. Thus, at a given speed v [m/s], the safety distance [m] is computed to equal (CC0 + CC1 • v). second 0.9 CC4 & CC5 Following thresholds: the thresholds which control the speed differences during the ‘Following’ state (In VISSIM, there are four driving modes: free driving, approaching, following, and braking). Smaller values of (CC4 & CC5) result in a more sensitive reaction of drivers to accelerations or decelerations of the preceding car, i.e. the vehicles are more tightly coupled. CC4 and CC5 are used for negative and positive speed differences, respectively. --- ± 0.35 Reduction factor for safety distance closed to stop line This reduction factor defines the vehicle behavior close to stop line at signalized intersections. This reduction factor is applied to the vehicle´s desired safety distance within a specific section started upstream the stop line and ended downstream the stop line. The specific distances upstream and downstream the stop line are predefined by the user. --- 0.60 Start upstream of stop line Distance upstream of the stop line of signalized intersection. Within this distance, the reduction factor is applied to the vehicle´s desired safety distance. m 100 Desired deceleration Desired deceleration is used as the maximum for the followings parameters: the deceleration caused by a desired speed decision; the deceleration in case of Stop & Go traffic, when closing up to a preceding vehicle; the deceleration towards an emergency stop position (route); and for co-operative braking. m/s2 -2.80 Table 4.2 Description of the selected VISSIM model parameters 52 4.3.2.2 Genetic Algorithm Once the important calibration parameters were determined, a Genetic Algorithm (GA) (Goldberg, 1989) process was used to calibrate the selected parameters. In this study, the range of each VISSIM parameter was established to be a set of discrete numbers instead of continuous range to simplify and accelerate the optimization process. Typically, the initial population is created randomly which means that all solutions have equal chances to fall into the population. However, this random selection does not guarantee the uniform covering of each parameter space. Rather, Latin Hypercube Sampling method (LHS) was used in this study to select the initial population. The first population in GA contained 10 individuals each of them represents one solution (6 values: one value for each VISSIM parameter). The fitness value, used to evaluate each solution, was the reciprocal of the summation of the correlations between the simulated and the field-measured rear-end conflicts at different TTC thresholds. The basic operators of GA, reproduction, crossover, mutation, and elitism were used to generate the next generation. The new generation has new 10 solutions (individuals) generated as follows: 8 solutions from crossover; one solution from mutation; and one solution from elitism (to keep the best solution from the previous generation). The 30 section-hours of the east and west approaches were used for the calibration process. 2100 simulation runs (70 generations X 10 individuals X 3 random seeds) were done. A computer program was developed to automate the process of calibration. The automation access of VISSIM was done using the VISSIM COM interface which enables user to access VISSIM through many scripting languages (such as Visual Basic or C++) (VISSIM 5.40-01 COM Interface Manual, 2011). However, the VISSIM COM interface does not cover all VISSIM 53 parameters. To overcome this problem, changing all the parameters values each time automatically was done by editing the text contents in the VISSIM (*.inp) files. After 70 generations, the optimum solution corresponded to the parameter’s values of (2.5 m) for CC0, (1.3 s) for CC1, (-0.25, +0.25) for CC4 & CC5, (0.75) for reduction factor for safety distance closed to stop line, (110 m) for start upstream of stop line, and (-2.8 m/s²) for desired deceleration. Figure 4.6 illustrates the flow chart of the applied GA procedure. The program codes used in this study is provided in Appendix D. More details about using GA in simulation models calibration can be found in (Park et al., 2004; Park et al., 2006; Park et al., 2005; Park et al. 2006). After the second-step calibration, the correlation between the field-measured and the simulated conflicts at all TTC thresholds was determined for the calibration dataset and the results will be discussed in section 4.3.3. Figure 4.6 Genetic Algorithm procedure flow chart 54 4.3.3 The Calibration Results The applied two-step calibration procedure enhanced the correlation between the simulated and the real rear-end conflicts at all TTC thresholds. Figure 4.7 shows the correlation between field-measured and simulated conflicts for the calibration dataset at the 21 threshold of TTC. Three scenarios were illustrated: default VISSIM, after the first calibration step, and after the second calibration step. For example, at the value of (1.5 seconds) for the TTC threshold, the correlation was increased from 0.46 at the default scenario to 0.65 after the first calibration, then to 0.72 after the second calibration. For the three scenarios, the results show that generally the higher the TTC threshold, the higher correlation between field-measured and simulated conflicts. This may be explained by the fact that, for a higher TTC threshold, there is a higher dependency on the exposure represented in traffic volume. Also, the enhancement in the correlation after the first calibration reflects the importance of the arrival type (platoon ratio) calibration and may be explained by the effect of the arrival type on the interactions between vehicles. The applied two-step calibration procedure enhanced the correlation between the simulated and the field-measured rear-end conflicts at all TTC thresholds, as shown in Figure 6. Three scenarios were illustrated: default VISSIM, after the first calibration step, and after the second calibration step. For example, at the value of (1.5 seconds) for the TTC threshold, the correlation was increased from 0.46 at the default scenario to 0.65 after the first calibration, then to 0.72 after the second calibration. For the three scenarios, the results show that generally the higher the TTC threshold, the higher the correlation between field-measured and simulated conflicts. This may be explained by the fact that, for a higher TTC threshold, there is a higher dependency on exposure represented by the traffic volume. Also, the enhancement in the correlation after the first calibration reflects 55 the importance of the arrival type (platoon ratio) calibration and may be explained by the effect of the arrival type on the interactions between vehicles. Figure 4.7 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (VISSIM) It is interesting to note that the first calibration step which controls the vehicle arrival rate and patterns and therefore will impact the opportunity for vehicle interactions (exposure) has a significant impact on the correlation. The second calibration step would mainly affect vehicle behavior in the simulation model by calibrating the values of the safety-related parameters. The second step calibration is shown to have smaller impact on the results. As well, the correlation between conflict rates (number of Conflicts/traffic Volume) was provided for each scenario. Figure 4.8 shows the correlation between field-measured and 56 simulated conflict rates for the calibration dataset at the 21 threshold of TTC. The results also show that there is no big difference in correlations based on conflict numbers and rates. Therefore, only the correlation between conflicts will be considered afterwards in the rest of this thesis. Figure 4.8 Correlation between simulated and field conflict rates at different TTC thresholds for calibration dataset (VISSIM) 4.4 Validation of the Proposed Procedure To validate the proposed two-step calibration procedure, the other two approaches in the first intersection, which were not included in the calibration (north and south approaches), were used as a validation dataset. Eighteen hours (9 hours X 2 approaches) were used to calculate the average delay time and 30 section-hours (i.e. sections 2, 4, 6, and 8 in Figure 3.2) were used to 57 calculate rear-end conflicts. First, arrival type and desired speed in the VISSIM model were calibrated to match the real conditions. Figure 4.9 shows the arrival types for the validation dataset in three scenarios: field-measured; default VISSIM; and VISSIM after the first calibration step. It can be noted that the real platoon ratio varies from 0.66 to 1.70. However, the default VISSIM model gives the platoon ratio varying from 1.24 to 1.61. Figure 4.10 illustrates the average delay time for the 18 hours. After the first calibration, the mean absolute percent error (MAPE) value for the average delay time decreased from 18.20% to 8.20%. Secondly, the calibrated values of the six calibration parameters were used in the VISSIM model and trajectory files were exported and processed in SSAM to get the simulated rear-end conflicts. Figure 4.9 Arrival types for validation dataset (VISSIM) 58 Figure 4.10 Average delays for validation dataset (VISSIM) The applied two-step calibration procedure enhanced the correlation between the simulated and the real rear-end conflicts for the validation dataset at all TTC thresholds, as shown in Figure 4.11. For example, at the value of (1.5 seconds) for TTC threshold, the correlation increased from 0.10 at default scenario to 0.50 after the first calibration, then to 0.66 after the second calibration. Similar to the calibration dataset results, the results show that generally the higher the TTC threshold, the higher correlation between real and simulated conflicts. Again, this may be explained by the high dependency on the exposure at high TTC thresholds. In addition, the validation dataset results prove the importance of the arrival type (platoon ratio) calibration for the enhancement in the correlation between real and simulated conflicts which may be explained again by the effect of arrival type on the interactions between vehicles. 59 Figure 4.11 Correlation between simulated and field conflicts at different TTC thresholds for validation dataset (VISSIM) 4.5 Conflicts Spatial Distribution For more investigation of the relationship between simulated and field-measured rear-end conflicts, besides the correlation, the locations of conflicts were identified. The locations of conflicts were determined for both calibration and validation datasets and compared with field-observed conflicts. For field-measured conflicts, the automated video-based computer vision technique can determine the position (x, y) of any event (e.g. minimum TTC) between two vehicles from their trajectories. For simulated conflicts, SSAM provides the positions, speeds, and directions of vehicles which are in conflict; then the conflict (minimum TTC) location can be extracted. A computer program was developed to adjust coordinates and project conflict 60 locations on the real image of the intersection for both simulated and field-measured conflicts. The spatial distribution of the rear-end conflicts was shown using heat maps. Heat maps represent the conflict intensity to highlight the hot spots (high conflict intensity locations). Figure 4.12 shows the heat maps of the calibration dataset for both simulated and field-measured conflicts. The heat maps show that there are major differences between real and simulated conflict locations for many sections. Figure 4.13 shows the heat maps of the validation dataset for both simulated and field-measured conflicts. Again, there are major differences between field-measured and simulated conflict locations for many sections. This may indicate that the correlation between the simulated and the real conflicts came from the similarity in the exposure (represented in traffic volumes and vehicles interaction) not from the similarity in the driving behavior. 61 Figure 4.12 Heat maps of calibration dataset (VISSIM) 62 Figure 4.13 Heat maps of validation dataset (VISSIM) 63 4.6 Summary In this chapter, the relationship between field-measured conflicts and simulated conflicts at an urban signalized intersection was investigated. Automated video-based computer vision techniques were used to extract vehicle trajectories and identify conflicts on all four approaches of the intersection. Conflict measures (e.g. TTC) and location were determined and compared with simulated conflicts from VISSIM using the SSAM tool. To increase the correlation between simulated and field conflicts, a two-step calibration procedure of VISSIM simulation model was proposed. In the first calibration step, the VISSIM model was calibrated to ensure that the simulation gives reasonable results of average delay times. Then, in the second calibration step, the Genetic Algorithm (GA) procedure was used to calibrate VISSIM parameters to enhance the correlation between simulated and field measured conflicts. The correlation between simulated and field-measured conflicts was investigated at 21 thresholds of TTC. Finally, heat maps were provided to compare field-measured and simulated conflict locations. The results showed that the calibration of the VISSIM model to match the existing traffic conditions (e.g. arrival pattern and platoon ratio) and to calibrate the driver behavior parameters is very important. This first calibration step has been generally ignored by researchers but shown in this study to have the most impact. In addition, it can be noted that if higher TTC thresholds values are used in the calibration, it will likely lead to good correlation between simulated and field-measured rear-end conflicts. However, this correlation is likely related to exposure dependency and does not demonstrate a real strong correlation. The spatial distribution showed that there are major differences in the locations between real and simulated conflicts at many sections which may indicate the simulation did not really capture the vehicles interaction (conflict) mechanism. 64 Chapter 5: PARAMICS – VISSIM Comparison This chapter contains six sections. The first section provides general information on the calibration framework of the simulation model (PARAMICS). The second section explains the establishment of the analyzed intersection using PARAMICS platform. The third section includes all details of the two-step calibration procedure of PARAMICS and the calibration results. The fourth section includes the details and the results of the validation of the proposed procedure. In the fifth section, a comparison will be performed between the two selected microsimulation models: PARAMICS and VISSIM. The sixth section provides a summary and discussions for this chapter. 5.1 PARAMICS Calibration Framework The SSAM approach can be used with trajectories extracted from several traffic microsimulation models. The main objective of this chapter is to investigate the relationship between field-measured and simulated conflicts at signalized intersections using another microsimulation model (PARAMICS). Field-measured data was obtained from the first intersection explained in Chapter 3. The applicability of the two-step calibration procedure applied to VISSIM in Chapter 4 was investigated and validated using PARAMICS. In the first calibration step, the PARAMICS model was calibrated to ensure that the simulation gives reasonable results of hourly average delay times. Then, in the second calibration step, the Genetic Algorithm (GA) procedure was used to calibrate PARAMICS parameters to enhance the correlation between simulated and field measured conflicts. 5.2 PARAMICS Platform The microsimulation model PARAMICS was used in this study to simulate the first intersection. PARAMICS depends on a psycho-physical car-following model based on the Fritzsche model 65 (Fritzsche and Daimler-benz Ag., 1994) which assumes that the driver can have one of five driving modes: Following-I, Following-II, Danger, Closing In, and Driving Freely. These modes are determined using six thresholds: Perception-Threshold Negative; Perception-Threshold Positive; Desired-Distance; Risky-Distance; Safe-Distance; and Braking-Distance (Fritzsche and Daimler-benz Ag., 1994; Panwai and Dia., 2005). In this study, in order to assess the intersections safety using simulation, the selected intersection has to be modeled accurately to match actual field conditions using PARAMICS Modeller. Modeller is the core network building post simulation tool within the PARAMICS suite of microsimulation tools (Quadstone PARAMICS User Manual - Modeller). First, the geometry of the intersection was extracted from aerial photo and then drawn in technical drawing. Once the geometry technical drawing was defined in the PARAMICS Modeller, all junctions, links, and zones were set up to match the real conditions. For each hour, the detailed traffic count extracted from videos was defined in PARAMICS using demand-editor (Quadstone Paramics User Guide, 2013) to represent all movements (i.e. through, left, right), and traffic composition (i.e. percentage of passenger cars, trucks, and buses). Public transit was defined in the PARAMICS model using the real number of public transit buses extracted from videos. Actual traffic signal settings were defined in PARAMICS for each hour using signalized junction with type actuated signal controller (ASC) (Quadstone Paramics User Guide, 2013; Quadstone PARAMICS User Manual - Actuated signal Controller). In the ASC editor, each phase has minimum green time, maximum green time, yellow time, and all red time. To represent the protected-permissive left turns, a detector was defined at each one of the 4 left-turn lanes. In each signal cycle, if the detector is occupied, a protected left turn phase will be called by ASC, otherwise the left turn will be permissive. As well, to avoid abnormal movements at junctions 66 when vehicles move from one link to another, next lanes were defined at each junction. In addition, some restriction rules were defined in the core network attributes and added to lane’s properties to make widening for left turns without abnormal lane changes. Figure 5.1 illustrates a screen shot of the developed PARAMICS model for the first intersection. Finally, visual inspection was done to check the simulation and to ensure that there is no abnormal movements of the simulated vehicles. Analyzer files were exported from PARAMICS Modeller and processed in PARAMICS Analyser (Quadstone PARAMICS User Manual - Analyser) to estimate the hourly average delay results for each approach. Trajectory files (*.trj files) were exported from PARAMICS Modeller and processed in SSAM to estimate the simulated rear-end conflicts at the 21 thresholds of TTC for the same 60 hours (30 hours for calibration dataset and 30 hours for validation dataset) pre-estimated using the automated video analysis in Chapter 3. The results of the average delay and the simulated rear-end conflicts will be discussed in the following sections. 67 Figure 5.1 Screen shot of the PARAMICS model for the first intersection 5.3 The PARAMICS Calibration Procedure The two-step calibration procedure proposed in Chapter 4 was applied to PARAMICS to enhance the correlation between the simulated and the field-measured rear-end conflicts using the calibration dataset (the east and west approaches (i.e. 72nd avenue) in the first intersection). 18 hours (9 hours X 2 approaches) were used to calculate average delay time and 30 section-hours (i.e. sections 1, 3, 5, 7 in Figure 3.2) were used to calculate rear-end conflicts. In the first calibration step, the PARAMICS model was calibrated to ensure that the simulation gives reasonable results in terms of average delay times. In the second calibration step, the PARAMICS driving behavior parameters were calibrated to enhance the correlation between the simulated and the field-measured rear-end conflicts. 68 5.3.1 First Step Calibration The main goal of this step is to match the simulated average delay times with the field-measured average delay times. In order to calibrate the average delay time, the arrival type was calibrated to match the field conditions. To match the field arrival type using PARAMICS Modeller, a dummy signalized junction was added before the four approaches of the analyzed intersection. Each dummy signalized junction has offset time (Δt) from the simulation time. By changing the offset time, the percentage of vehicles (P) arriving on green time of the analyzed intersection can be changed. The offset time, which gives a value for P that matches the field condition, was obtained for each approach each hour. As well, a bypass at each approach was added to represent the percentage of vehicles (X %) arriving out of platoon. The percentage X was measured from the field data as average for each hour. The percentage X was defined in PARAMICS by defining two routes: the bypass route which has X% of the vehicles; and the dummy signalized junction route which has (1-X) % of vehicles. Figure 5.2 illustrates the process of arrival type calibration. Figure 5.2 Arrival type calibration process (PARAMICS) 69 Figure 5.3 shows the different arrival types for 18 hours (9 hours X 2 approaches). It can be noted that the field platoon ratio varies from 0.53 to 1.46; however, the default VISSIM model gives the platoon ratio varying from 0.46 to 1.06. Figure 5.4 illustrates the average delay time for the 18 hours. After the first calibration, the mean absolute percent error (MAPE) value for the average delay time was decreased from 30.20% to 7.60%. This shows the importance of the arrival type calibration at signalized intersections when estimating average delay time using microsimulation models. Figure 5.3 Arrival types for calibration dataset (PARAMICS) 70 Figure 5.4 Average delays for calibration dataset (PARAMICS) After the first-step calibration, the trajectory files were exported from PARAMICS and imported in SSAM in order to estimate the simulated rear-end conflicts. Simulated rear-end conflicts for the selected 30 section-hours were estimated at the 21 thresholds of TTC. The correlation between the field-measured and simulated conflicts at all TTC thresholds was determined and the results will be discussed later in section 5.3.3. 5.3.2 Second Step Calibration The main purpose of this step is to enhance the correlation between field-measured and simulated conflicts by calibrating PARAMICS parameters. Three car-following-model parameters were selected: Mean target headway (MTH), Mean driver reaction time (MRT), and Minimum gap (G). The three parameters affect the desired safety distance between each couple of consecutive vehicles in PARAMICS, and can be considered as the safety-related parameters at signalized intersections. Table 5.1 provides more details of these parameters in the context of the 71 driving behavior in PARAMICS. More description of the selected parameters can be found in (Quadstone PARAMICS User Manual - Modeller). Parameter Description Unit Default Mean Target Headway (MTH) MTH specifies the global mean target headway, in seconds, between a vehicle and a following vehicle. This will not necessarily be equal to the mean measured headway: the relationship between target and actual depends on traffic flow levels, driver behavior and several other factors. second 1.00 Mean Driver Reaction Time (MRT) The mean reaction time of each driver, in seconds. The value is associated with the lag in time between a change in speed of the preceding vehicle and the following vehicles reaction to the change. second 0.90 Minimum Gap (G) The minimum gap between stationary vehicles in a queue. m 2.00 Table 5.1 Description of the selected PARAMICS model parameters Subsequently, a Genetic Algorithm (Goldberg, 1989) technique was applied to estimate the best values of these parameters which give the highest correlation between simulated and field-measured rear-end conflicts at different TTC thresholds. The range of each PARAMICS parameter was established to be a set of discrete numbers instead of continuous range to simplify and accelerate the optimization process. The range of each parameter was determined based on previous studies (Saleem et al. 2014; Park et al., 2004). Latin Hypercube Sampling method (LHS) was used to select the initial population. The first population in GA contained 10 individuals each of them represents one solution (3 values: one value for each PARAMICS parameter). The fitness value, used to evaluate each solution, was the reciprocal of the summation of the correlations between the simulated and the field-measured rear-end conflicts at different TTC thresholds. The basic operators of GA, reproduction, crossover, mutation, and elitism were used to generate the next generation. The new generation has new 10 solutions 72 (individuals) generated as follows: 8 solutions from crossover; one solution from mutation; and one solution from elitism (to keep the best solution from the previous generation). 1200 simulation runs (40 generations X 10 individuals X 3 random seeds) were done to obtain the optimum solution. A computer program was developed to automate the process of calibration. The automation access of PARAMICS Modeller was done using the processor-cmd application (Quadstone PARAMICS User Manual) which enables user to access PARAMICS Modeller through the command prompt. For each run, changing all the three parameters values automatically was done by editing the text contents in the PARAMICS network configuration file. After 40 generations of the applied GA, the second calibration step ended up with the best values of the three PARAMICS parameters as following: 0.8 seconds for MTH; 1.6 seconds for MRT; and 1.50 m for G. The program codes used in this study is provided in Appendix D. More details about using GA in simulation models calibration can be found in (Park et al., 2004; Park et al., 2005; Park et al., 2006; Park et al. 2006). After the second-step calibration, the correlation between the field-measured and the simulated conflicts at all TTC thresholds was determined for the calibration dataset and the results will be discussed later in the following section. 5.3.3 The Calibration Results The applied two-step calibration procedure enhanced the correlation between the simulated and the field-measured rear-end conflicts at all TTC thresholds. Figure 5.5 shows that the correlation between simulated and field-measured conflicts has increased at all TTC thresholds after the two calibration steps. For example, at TTC of 1.5 seconds, the correlation increased from 0.20 (default) to 0.55 (after first calibration), and then to 0.75 (after the second calibration). 73 Figure 5.5 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (PARAMICS) For the three scenarios shown in Figure 5.5, the results show that generally the higher the TTC threshold, the higher correlation between field-measured and simulated conflicts. This may be explained by the fact that, for a higher TTC threshold, there is a higher dependency on the exposure as represented by traffic volume. Also, it can be noted that the first calibration step which controls the vehicle arrival rate and patterns and therefore will impact the opportunity for vehicle interactions (exposure) has a significant impact on the correlation. The second calibration step would mainly affect vehicle behavior in the simulation model by calibrating the values of the safety-related parameters. The second step calibration is shown to have smaller impact on the results. 74 5.4 Validation of the Proposed Procedure The validation dataset (north and south approaches which were not included in the calibration process) was used to validate the two-step calibration procedure. Eighteen hours (9 hours X 2 approaches) were used to calculate the average delay time and 30 section-hours (i.e. sections 2, 4, 6, and 8 in Figure 3.2) were used to calculate rear-end conflicts. The two-step proposed procedure was applied on the PARAMICS Modeller of the validation dataset. In the first step, arrival type in the PARAMICS Modeller was calibrated to match the field conditions and the hourly average delay results were estimated. The simulated hourly average delay time values were calculated before and after the first calibration step. Figure 5.6 shows the arrival types for the validation dataset in three scenarios: field-measured; default PARAMICS; and PARAMICS after the first calibration step. It can be noted that the field platoon ratio varies from 0.66 to 1.70. However, the default PARAMICS model gives the platoon ratio varying from 0.53 to 1.17. Figure 5.7 illustrates the average delay times for the validation dataset. After the first calibration, the mean absolute percent error (MAPE) value for the average delay time decreased from 43.40% to 6.70%. 75 Figure 5.6 Arrival types for validation dataset (PARAMICS) Figure 5.7 Average delays for validation dataset (PARAMICS) 76 In the second step, instead of the default values, the calibrated values of the three calibration parameters, which were obtained from the calibration dataset, were used in the PARAMICS Modeller. Simulated rear-end conflicts were estimated for the validation dataset at the 21 thresholds of TTC. The correlation between the field-measured and the simulated conflicts was determined at all TTC thresholds for three scenarios: default PARAMICS; PARAMICS after first calibration step; PARAMICS after second calibration step (using calibrated values of PARAMICS parameters obtained from the calibration dataset). Figure 5.8 shows that the applied two-step calibration procedure enhanced the correlation between the simulated and the field-measured rear-end conflicts for the validation dataset at all TTC thresholds. For example, at the value of (1.5 seconds) for TTC threshold, the correlation increased from 0.34 at default scenario to 0.52 after the first calibration, then to 0.82 after the second calibration. 77 Figure 5.8 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (PARAMICS) Similar to the calibration dataset results, the validation results show that generally the higher the TTC threshold, the higher correlation between field-measured and simulated conflicts. Again, this may be explained by the high dependency on the exposure at high TTC thresholds. In addition, the validation dataset results confirm the effect of both calibration steps on enhancing the correlation between simulated and field-measured conflicts. 5.5 PARAMICS - VISSIM Comparison In this section, a comparison will be performed between the two selected microsimulation models: PARAMICS and VISSIM. The comparison includes three sections: 1) car-following model and safety-related parameters; 2) correlation between simulated and field-measured 78 conflicts; and 3) conflict spatial distribution. The PARAMICS simulation model results obtained earlier in this chapter will be compared to the results obtained from Chapter 4, in which the same signalized intersection was analyzed using the VISSIM simulation model. 5.5.1 Car-Following-Model and Safety-Related Parameters PARAMICS and VISSIM depend on two different car following models. However, both models are psycho-physical car following models, in which there are different driving modes determined by some thresholds. In the second calibration step in this chapter, three PARAMICS parameters were calibrated and considered as safety-related parameters. On the other hand, in Chapter 4, six VISSIM parameters were used as safety-related parameters in the calibration process. Table 5.2 provides a brief summary of each microsimulation model in terms of the car following model used, and the model parameters used in the calibration process to enhance the correlation between simulated and field-measured rear-end conflicts. 79 Table 5.2 PARAMICS and VISSIM model parameters * (Quadstone Paramics User Guide, 2013) ** (VISSIM 5.30-05 User Manual, 2011) *** (Fritzsche and Daimler-benz Ag., 1994; Panwai and Dia., 2005) **** (Wiedemann, 1974; Wiedemann, 1991) 80 5.5.2 Correlation between Simulated and Field-measured Conflicts The correlations between the simulated and the field-measured conflicts from both PARAMICS and VISSIM were compared at three scenarios: default model; after first calibration step; and after second calibration step. Figure 5.9 shows the correlations for the calibration dataset, and Figure 5.10 shows the correlations for the validation dataset. The results show that default PARAMICS and default VISSIM give poor correlation with the field-measured data, and therefore using these models without a proper calibration should be avoided. For both microsimulation models, it can be noted that the first calibration step has a significant impact on the results. This confirms the effect of the first calibration step (arrival type calibration) which impacts the opportunity for vehicle interactions (exposure) at signalized intersections. Also, it is obvious that the higher the TTC threshold, the higher correlation between field-measured and simulated conflicts. This may be explained by the fact that, for a higher TTC threshold, there is a higher dependency on the exposure represented by traffic volume. 81 Figure 5.9 Correlation between simulated and field conflicts at different TTC thresholds for calibration dataset (VISSIM - PARAMICS) 82 Figure 5.10 Correlation between simulated and field conflicts at different TTC thresholds for validation dataset (VISSIM - PARAMICS) For further investigation of the calibration results for both PARAMICS and VISSIM, linear regression models were developed to relate the number of the field-measured conflicts to the number of the simulated conflicts at a TTC threshold of (1.50 seconds). The TTC threshold of (1.50 seconds) is the default of the SSAM tool and is commonly used by researchers to define rear-end conflicts. Figure 5.11 shows conflict regression models for PARAMICS and VISSIM for three cases: default model; after the first calibration step; and after the second calibration step. The results show that the coefficient of correlation (R2) of the linear model increased from 0.13 using the default PARAMICS parameters to 0.40 after the first calibration; and from 0.14 using the default VISSIM parameters to 0.45 after the first calibration. This emphasizes the 83 importance of the first calibration step. The correlation coefficient increased to 0.63 after the second calibration step of PARAMICS and to 0.55 after the second calibration step of VISSIM. Although PARAMICS and VISSIM give close correlations after the second calibration step, the results show that PARAMICS generally overestimates the number of conflicts while VISSIM underestimates them. This means that even with the high correlation, the microsimulation models may have not captured the real conflicts mechanism. Figure 5.11 Regression models using simulated and field-measured conflicts (TTC = 1.5 seconds) 84 5.5.3 Conflict Spatial Distribution For more investigation of the relationship between simulated and field-measured rear-end conflicts beyond the correlation, the spatial distributions of conflicts were compared. The locations of conflicts were determined after the second calibration step; then compared with field-measured conflicts. The spatial distribution of the rear-end conflicts was shown using heat maps for field-measured conflicts, PARAMICS simulated conflicts, and VISSIM simulated conflicts. Figure 5.12 and Figure 5.13 show the heat maps of the rear-end conflicts for the calibration dataset and the validation dataset, respectively. The heat maps show that there are major differences between field-measured and simulated conflict locations for the approaches for both simulation models. This indicates that despite the good correlation obtained from the calibration process, both PARAMICS and VISSIM may have failed to well capture the actual conflict occurrence mechanism. One explanation for the correlation between the simulated and the real conflicts is the relationship between conflicts and exposure (represented in traffic volumes and vehicles interaction). 85 Figure 5.12 Heat maps of calibration dataset (Field conflicts – PARAMICS conflicts – VISSIM conflicts) 86 Figure 5.13 Heat maps of validation dataset (Field conflicts – PARAMICS conflicts – VISSIM conflicts) 5.6 Summary This chapter investigated the relationship between field-measured conflicts and simulated conflicts from the microsimulation model (PARAMICS) using the SSAM tool. The applicability of the two-step calibration procedure applied to VISSIM in Chapter 4 was investigated and validated using PARAMICS. In the first calibration step, the PARAMICS model was calibrated to ensure that the simulation gives reasonable results of hourly average delay times. Then, in the second calibration step, the Genetic Algorithm (GA) procedure was used to calibrate 87 PARAMICS parameters to enhance the correlation between simulated and field-measured conflicts. Three PARAMICS parameters were included in the calibration process: Mean target headway, Mean driver reaction time, and Minimum gap. The correlation between simulated and field-measured conflicts was investigated at 21 thresholds of TTC. Finally, a comparison was performed between the two selected microsimulation models: PARAMICS and VISSIM. The comparison included three aspects: 1) car-following model and safety-related parameters; 2) correlation between simulated and field-measured conflicts; and 3) conflict spatial distribution. The results showed that the two steps of PARAMICS calibration are very important. The exposure effect is likely included in the first calibration step which controls the vehicle arrival rate and therefore will impact the opportunity for interacting. On the other hand, the second calibration step would mainly affect vehicle behavior by controlling the values of the safety-related parameters in the PARAMICS simulation model. For both simulation models: PARAMICS and VISSIM, the results show that default simulation model gives poor correlation with the field-measured data, and therefore using simulation models without a proper calibration should be avoided. For both microsimulation models, it can be noted that the first calibration step has a significant impact on the results. This emphasizes the effect of the first calibration step (arrival type calibration) which impacts the opportunity for vehicle interactions (exposure) at signalized intersections. Also, it is obvious that the higher the TTC threshold, the higher correlation between field-measured and simulated conflicts. This may be explained by the fact that, for a higher TTC threshold, there is a higher dependency on the exposure represented in traffic volume. Linear regression results show that the coefficient of correlation (R2) of the linear model increased from 0.13 using the default PARAMICS parameters to 0.40 after the first calibration; 88 and from 0.14 using the default VISSIM parameters to 0.45 after the first calibration. This emphasizes the importance of the first calibration step. The correlation coefficient increased to 0.63 after the second calibration step of PARAMICS and to 0.55 after the second calibration step of VISSIM. Although PARAMICS and VISSIM give close correlations after the second calibration step, the results show that PARAMICS overestimates the number of conflicts while VISSIM underestimates them. This means that even with the high correlation, the microsimulation models do not capture the real conflicts mechanism. The conflict spatial distributions, represented by the heat maps, show that there are major differences between field-measured and simulated conflict locations for all approaches for both simulation models. This indicates that despite the good correlation obtained from the calibration process, both PARAMICS and VISSIM failed to capture the actual conflict occurrence mechanism well. 89 Chapter 6: Transferability of Simulation Models This chapter contains six sections. The first section introduces the need of the simulation models transferability investigation. The second section provides information on the transferability approaches. The third chapter presents the transferability investigation of the simulation model through five different scenarios. The fourth section includes an investigation of the transferability of individual parameters in the simulation model. The fifth section includes the spatial distribution of the conflicts using heat maps. The sixth section provides a summary and discussions for this chapter. 6.1 The Need of Transferability Investigation The previous two-steps calibration procedure proposed in this study needs field-measured data to be collected, and field conflicts to be identified. Furthermore, the genetic algorithm procedure may take long time to get the best solution after a number of generations. So, the calibration process can be relatively complicated, time consuming, and need real traffic conflict data to be collected. In addition, the main advantage of using the microsimulation models and SSAM is the evaluation of the safety of changes to design and traffic control that have not been implemented. Therefore, the transfer of the calibrated parameters of simulation models for use in safety studies between different locations (with comparable traffic conditions and geometric characteristics) can help in eliminating or reducing the need for the complex calibration process. Therefore, there is a need to investigate and confirm the transferability of the calibrated simulation model parameters for use at different sites. The main objective of this chapter is to investigate the transferability of calibrated parameters of microscopic simulation models for safety analysis between two different sites. This chapter investigates whether the transferred parameters, when applied to other sites, give 90 reasonable results in terms of the relationship (correlation) between the field-measured and the simulated rear-end conflicts. 6.2 Transferability Approaches There are two main approaches for investigating the transferability of a certain model (Bowman et al., 2013; Sawalha and Sayed, 2006):  Application-based approach: This is the more direct and easier approach; in which, the model parameters are calibrated by using data from one location (the base context) and applied with no change to data in another location (the application context) to assess how well the calibrated model predicts in the other location. Although this alternative is easier, this approach generally tests the transferability of model as a whole without allowing an examination of which specific parameters are transferable (Sikder et al., 2014).  Estimation-based approach: This is the more desirable transfer approach; in which the model parameters are calibrated by using data from one location, and recalibrated by using data from another location. The model transferability is investigated by identifying whether the calibrated parameters values are different between the two locations. The main advantage of this approach is that one can test whether each parameter in a model is transferable (Sikder et al, 2014). Both approaches are used in this study to investigate the transferability of the calibrated parameters of the simulation model. 6.3 Simulation Model Transferability Investigation The data from the second intersection (i.e. 72nd Avenue and 132nd Street) was used to investigate transferability of the calibrated simulation model. The VISSIM model was used as an example of the traffic microsimulation models. The correlation between simulated and field-measured 91 conflicts was calculated for VISSIM model of the second intersection for five scenarios. In the first scenario, no parameter calibration was undertaken for the second intersection and the default values of the VISSIM parameters were used. In the second scenario, the calibrated values of the VISSIM parameters obtained from the first intersection (in Chapter 4) were used for the second intersection. In the third scenario, only the first calibration step was applied, and the default values of the VISSIM parameters were used for the second intersection. In the fourth scenario, only the first calibration step was applied, and the calibrated values of the VISSIM parameters obtained from the first intersection (in Chapter 4) were used for the second intersection. In the fifth scenario, both calibration steps were applied and new calibrated values of the VISSIM parameters were determined from the second intersection. Table 6.1 summarizes, for each scenario, the used values of VISSIM parameters, and which calibration step was applied. Figure 8 shows both the correlation between simulated and field-measured conflicts and the correlation between simulated and field-measured conflict rates at all TTC thresholds for the five scenarios. Scenario # VISSIM Parameters First Calibration Step (Arrival Type and Desired Speed) Second Calibration Step (GA) CC0 CC1 CC4&CC5 Reduction Factor For Safety Distance Close to Stop Line Start Upstream before Stop Line Desired Deceleration Description 1 1.5 0.9 ±0.35 0.60 100 -2.80 Default No No 2 2.5 1.3 ±0.25 0.75 110 -2.80 Transferred No No 3 1.5 0.9 ±0.35 0.60 100 -2.80 Default Yes No 4 2.5 1.3 ±0.25 0.75 110 -2.80 Transferred Yes No 5 2.1 1.3 ±1.1 0.6 100 -2.80 Calibrated Yes Yes Table 6.1 Five scenarios at the second intersection for transferability investigation 92 6.3.1 Scenario 1 In this scenario the default VISSIM parameters were used for modeling the second intersection without calibration. The correlation between field-measured and simulated conflicts was estimated. The results in Figure 6.1 indicate that when using the default values of VISSIM parameters, the field-measured and simulated conflicts are not correlated. Thus, using simulation models and SSAM to evaluate safety without proper calibration of simulation models can lead to biased results and should be avoided. 6.3.2 Scenario 2 In this scenario the first and the second calibration steps were ignored, and the calibrated VISSIM parameters from the first intersection were transferred directly to the second intersection without recalibration. The correlation between field-measured and simulated conflicts was estimated. The results in Figure 6.1 indicate that using the transferred values of VISSIM parameters can have a significant improvement in the correlation between field-measured and simulated conflicts at all TTC thresholds. However, the correlation coefficient still has a maximum value of (0.37) which is considered small and needs to be enhanced. 6.3.3 Scenario 3 In this scenario the default VISSIM parameters were used for modeling the second intersection, and the first calibration step (arrival type and desired speed) only was applied. The correlation between field-measured and simulated conflicts was estimated. The results in Figure 6.1 indicate that the correlation between field-measured and simulated conflicts has improved at all TTC thresholds comparing to the scenarios 1 and 2. This clearly emphasizes the need to do the first step calibration to match vehicle arrival types and desired speed. However, the correlation still 93 has a maximum value of (0.56) at TTC threshold of (3 seconds). Therefore, the correlation needs to be improved by using different parameter values than defaults. 6.3.4 Scenario 4 In this scenario the first calibration step was applied for the second intersection and the calibrated VISSIM parameters from the first intersection were transferred directly to the second intersection without recalibration. The correlation between field-measured and simulated conflicts was estimated. The results in Figure 6.1 show that the correlation between simulated and field-measured conflicts was further enhanced at all TTC thresholds when the transferred values of the parameters were used instead of default values. The correlation has values range from (0.52) to (0.75). Thus, if the second calibration step (local parameter optimization) was ignored, the simulated conflicts can still have a good correlation with the field-measured conflict. This confirms that the main VISSIM parameters that affect conflicts are transferable for use at similar intersections. 6.3.5 Scenario 5 In this scenario the first and the second calibration steps were conducted using the second intersection field-measured conflicts. In the second calibration step, the Genetic Algorithm (GA) (Goldberg, 1989) process (explained in section 4.3.2.2) was applied to calibrate the most six important parameters identified from the sensitivity analysis. The 23 hours of the east and west approaches at the second intersection were used for the calibration process. 600 simulation runs (20 generations X 10 individuals X 3 random seeds) were done. A computer program was developed to automate the process of calibration. After 20 generations, the optimum solution corresponded to the parameter’s values of (2.1 m) for CC0, (1.3 seconds) for CC1, (-1.1, +1.1) for CC4 & CC5, (0.6) for reduction factor for safety distance 94 closed to stop line, (100 m) for start upstream of stop line, and (-2.8 m/s²) for desired deceleration. As shown in Figure 6.1, the correlation between simulated and field-measured conflicts is higher for Scenario 5 (complete calibration process using conflict data) than all other scenarios. The correlation has a maximum value of (0.81). This enhancement in correlation is expected due to the local calibration process. The difference in correlation between scenario 4 and scenario 5 is evident but not very large. Therefore, although local calibration of model parameters is important, using calibrated parameters transferred from similar intersections can lead to good results. Figure 6.1 Correlation between simulated and field-measured conflicts/conflict rates at the second intersection for scenarios (1), (2), (3), (4) and (5) 95 6.4 Transferability of Individual Parameters For further examination of the transferability of the VISSIM model calibrated parameters, the transferability of each parameter was investigated by comparing its calibrated value between the two selected intersections. From sensitivity analysis, there are 6 parameters which have the biggest effect on the number of the simulated conflicts. The values of these parameters were calibrated using GA procedure for both intersections (full calibration process). The percentage of change for each parameter was determined by the following equation: % 𝑜𝑓 𝑐ℎ𝑎𝑛𝑔𝑒 =|𝑉1 − 𝑉2|𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑣𝑎𝑙𝑢𝑒 − 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 𝑣𝑎𝑙𝑢𝑒𝑥 100 Where: V1: The calibrated value of the parameter from the first intersection V2: The calibrated value of the parameter from the second intersection Maximum value: The maximum value of the parameter Minimum value: The minimum value of the parameter The maximum and the minimum values of the parameters were assumed based on previous studies (Cunto and Saccomanno, 2008; Park et al., 2006) and VISSIM user Manual (VISSIM 5.30 User Manual, 2011). Table 6.2 provides the default values, range, and calibrated values of VISSIM parameters from each intersection. 96 Table 6.2 Calibrated VISSIM parameters from the first and the second intersections As shown in Table 6.2, when the calibration process was applied at the two intersections, there is no change in the calibrated values of the parameters (CC1) and (desired deceleration); which means that these two parameters are directly transferable. The parameters (CC0), (reduction factor for safety distance closed to stop line), and (start upstream of stop line) have percentage of changes 20%, 16.67%, and 5% respectively. The low values of the percentage of change of these three parameters mean that they are also transferable to some degree. However, they can have better values if a local calibration is applied. The parameters (CC4 & CC5) have percentage of change equals 57.05%. This high change means that this parameter is not transferable and needs a new calibration to enhance the results. This indicates that, if conflict data is available for use in the calibration, one to three parameters can be used instead of 6 parameters. This can reduce the search space when using GA process and subsequently decreases the number of generations required to reach the optimum solution. Thus, this can make the second step of the calibration process explicitly easier and faster. 6.5 Conflict Spatial Distribution For more investigation of the relationship between simulated and field-measured rear-end conflicts beyond the correlation between simulated and field conflicts, the spatial distributions of 97 conflicts were compared. The locations of conflicts were determined for scenarios 4 and 5; then compared with field-measured conflicts. For field-measured conflicts, the automated video-based computer vision technique can determine the position (x, y) of any event (e.g. minimum TTC) between two vehicles from their trajectories. For simulated conflicts, SSAM provides the positions, speeds, and directions of vehicles which are in conflict; then the conflict (minimum TTC) locations were calculated. A computer program was developed to adjust coordinates and project conflict locations on the real image of the intersection for both simulated and field-measured conflicts. The spatial distribution of the rear-end conflicts was shown using heat maps that represent the conflict intensity. Figure 6.2 shows the heat maps of the rear-end conflicts at the second intersection. The heat maps are shown for both field-measured conflicts and simulated conflicts for scenarios 4 and 5. The heat maps show that there are major differences between field-measured and simulated conflict locations for the approaches. This indicates that despite the good correlation obtained from the calibration process, the simulation model and SSAM did not capture the actual conflict occurrence mechanism well. One explanation for the correlation between the simulated and the real conflicts is the relationship between conflicts and exposure (represented in traffic volumes and vehicles interaction). 98 Figure 6.2 Spatial distribution of the field-measured conflicts, simulated conflicts with transferred parameters (Scenario 4), and simulated conflicts with calibrated parameters (Scenario 5) 99 6.6 Summary The main objective of this chapter was to investigate the transferability of calibrated parameters of microscopic simulation models for safety analysis between two different sites. The main purpose is to investigate whether the transferred parameters, when applied to other sites, give reasonable results in terms of the relationship (correlation) between the field-measured and simulated rear-end conflicts. The VISSIM model was used as an example of the traffic microsimulation models. The transferability of the calibrated simulation model was investigated by estimating the correlation between field-measured and simulated conflicts for VISSIM model with default and transferred parameters and with and without the calibration process. In addition, the calibrated values of the VISSIM parameters from the two intersections (shown in Chapter 2) were compared and their transferability was determined. Finally, the spatial distributions of the field-measured and the simulated conflicts were compared through heat maps. The results showed that the two steps of calibration (matching the existing traffic conditions and calibrating the driver behavior parameters) are very important. The exposure effect is likely included in the first calibration step which controls vehicle arrival rate and patterns and therefore will impact the opportunity for interacting (exposure). This first calibration step has been generally ignored by researchers but shown in this study to have significant impact. On the other hand, the second calibration step would mainly affect vehicle behavior by controlling the values of the safety-related parameters in the simulation model. In addition, if only the first calibration step was applied (ignoring the second calibration step) and the transferred values of the calibrated parameters are used instead of the default values, the simulated conflicts will have a relatively good correlation with the field-measured conflict. However, this correlation will slightly increase if a local second calibration step is applied. 100 The transferability of each parameter was examined by comparing its calibrated value between the two selected intersections. The results showed that there is no change in the calibrated values of the parameters (CC1) and (desired deceleration); which means that these two parameters are directly transferable. Also, The parameters (CC0), (reduction factor for safety distance closed to stop line), and (start upstream of stop line) changed by 20%, 16.67%, and 5% respectively. The low values of the percentage of change of these three parameters mean that they are also transferable to some degree. The parameters (CC4 & CC5) have percentage of change equals 57.05%. This high change means that these parameters are not transferable and may need a new calibration to enhance the results. This indicates that, if conflict data is available for use in the calibration, one to three parameters can be used instead of 6 parameters. This can make the second step of the calibration process easier and faster. A comparison of the spatial distributions of simulated and field-measured conflicts showed major differences indicating that despite the high correlation, the simulation did not capture the vehicles interaction (conflict occurrence) mechanism. More work is still needed to confirm that simulated conflicts provide safety measures beyond what can be expected from exposure. 101 Chapter 7: Summary and Conclusions 7.1 Summary In this thesis, the relationship between field-measured traffic conflicts and simulated traffic conflicts at signalized intersections was investigated. Two signalized intersections at the same corridor in Surrey, BC were selected for the analysis. At both intersections, automated video-based computer vision techniques were used to extract vehicle trajectories and identify field-measured rear-end conflicts. The simulated rear-end conflicts were estimated using SSAM tool by analyzing the vehicles trajectories extracted from two different microsimulation models (VISSIM and PARAMICS). At the first intersection, to increase the correlation between simulated and field-measured conflicts, a two-step calibration procedure of VISSIM simulation models was proposed and validated. In the first calibration step, the vehicle arrival type in VISSIM model was calibrated to ensure that the simulation gives reasonable results of average delay times. Then, in the second calibration step, the Genetic Algorithm (GA) procedure was used to calibrate VISSIM parameters to enhance the correlation between simulated and field measured conflicts. Six VISSIM parameters were included in the calibration process. The correlation between simulated and field-measured conflicts was investigated at 21 thresholds of time-to-collision (TTC). Furthermore, the applicability of the proposed calibration procedure applied to VISSIM was investigated using PARAMICS. Three PARAMICS parameters were included in the second calibration step. As well, the heat maps were provided to compare field-measured and simulated conflict locations of both simulation models. Finally, the results obtained from PARAMICS were compared to results obtained from VISSIM. The comparison included three aspects: 1) the car-102 following model and safety-related parameters; 2) the correlation between simulated and field-measured conflicts; and 3) the conflict spatial distributions. At the second intersection, the transferability of the calibrated simulation model was investigated by estimating the correlation between field-measured and simulated conflicts for VISSIM model with default and transferred parameters and with and without the calibration process. In addition, the calibrated values of the VISSIM parameters from the two intersections were compared and their transferability was determined. As well, the spatial distributions of the field-measured and the simulated conflicts were compared through heat maps. 7.2 Conclusions For both simulation models: PARAMICS and VISSIM, the analysis of traffic conflict results showed that the default simulation model parameters give poor correlation with the field-measured data, and therefore using simulation models in estimating traffic conflicts without a proper calibration should be avoided. The proposed two-step calibration procedure enhanced the correlation between the simulated and the field-measured conflicts at different thresholds of time-to-collision (TTC). The results showed that the two steps of calibration (matching the existing traffic conditions and calibrating the driver behavior parameters) are very important. The exposure effect is likely included in the first calibration step which controls vehicle arrival rate and patterns and therefore will impact the opportunity for interacting (exposure). This first calibration step has been generally ignored by researchers but shown in this study to have significant impact. On the other hand, the second calibration step would mainly affect vehicle behavior by controlling the values of the safety-related parameters in the simulation model. 103 In addition, it can be noted that if higher TTC thresholds values are used in the calibration, it will likely lead to good correlation between simulated and field-measured rear-end conflicts. However, this correlation is likely related to exposure dependency and does not demonstrate a real strong correlation. Linear regression results show that the coefficient of correlation (R2) of the linear model increased from 0.14 using the default VISSIM parameters to 0.45 after the first calibration; and from 0.13 using the default PARAMICS parameters to 0.40 after the first calibration. This emphasizes the importance of the first calibration step. The correlation coefficient increased to 0.55 after the second calibration step of VISSIM and to 0.63 after the second calibration step of PARAMICS. Although PARAMICS and VISSIM give close correlations after the second calibration step, the results show that PARAMICS overestimates the number of conflicts while VISSIM underestimates them. This means that even with the high correlation with the field conflicts, the microsimulation models do not capture the real conflicts mechanism. The conflict spatial distributions, represented by the heat maps, show that there are major differences between field-measured and simulated conflict locations for all approaches for both simulation models. This indicates that despite the good correlation obtained from the calibration process, both PARAMICS and VISSIM failed to capture the actual conflict occurrence mechanism well. More work is still needed to confirm that simulated conflicts provide safety measures beyond what can be expected from exposure. The investigation of the transferability of the calibrated simulation model VISSIM between different sites emphasizes the importance of the two proposed steps of calibration. In addition, if only the first calibration step was applied (ignoring the second calibration step) and the transferred values of the calibrated parameters are used instead of the default values, the 104 simulated conflicts will have a relatively good correlation with the field-measured conflict. However, this correlation will slightly increase if a local second calibration step is applied. The transferability of each VISSIM parameter was examined by comparing its calibrated value between the two selected intersections. The results showed that there is no change in the calibrated values of the parameters (CC1) and (desired deceleration); which means that these two parameters are directly transferable. Also, The parameters (CC0), (reduction factor for safety distance closed to stop line), and (start upstream of stop line) changed by 20%, 16.67%, and 5% respectively. The low values of the percentage of change of these three parameters mean that they are also transferable to some degree. The parameters (CC4 & CC5) have percentage of change equals 57.05%. This high change means that these parameters are not transferable and may need a new calibration to enhance the results. This indicates that, if conflict data is available for use in the calibration, one to three parameters can be used instead of 6 parameters. This can make the second step of the calibration process easier and faster. 7.3 Recommendations for Future Research One of the limitations of this study is that the analysis included only two signalized intersections. Therefore, it is recommended that a larger study with several intersections be conducted to confirm the results and the values of the calibrated parameters of both simulation models VISSIM and PARAMICS. In addition, this study was concerned with one type of conflict (rear-end), one safety measure indicator (i.e. TTC), and at one road facility (signalized intersections). Other conflict types (e.g. lane-change and crossing), other safety measure indicators (e.g. PET, DR, MaxS, and DeltaS), and other road facilities (e.g. un-signalized intersections and freeways) may be discussed in future studies. Finally, more work is needed on investigating the relationship 105 between collisions and both field-measured and simulated conflicts (Sacchi et al., 2013; El-Basyouny and Sayed, 2013). 106 Bibliography A Policy on Geometric Design of Highways and Streets,5th ed. AASHTO. Washington, D.C., 2004. Amundsen, F., and C. Hydén. Proceedings of First Workshop on traffic Conflicts. Oslo, s.n.: Institute of Economics, 1977. Archer, Jeffery. Indicators for traffic safety assessment and prediction and their application in micro-simulation modelling: A study of urban and suburban intersections. Doctoral Dissertation, Royal Institute of Technology, 2005. —. Methods for the assessment and prediction of traffic safety at urban intersections and their application in micro-simulation modelling. Royal Institute of Technology, 2004. Atev, Stefan, Hemanth Arumugam, Osama Masoud, Ravi Janardan, and Nikolaos P. Papanikolopoulos. \"A vision-based approach to collision prediction at traffic intersections.\" Intelligent Transportation Systems, IEEE Transactions on 6, no. 4, 2005: 416-423. Autey, J., Tarek Sayed, and M. Zaki. \"Safety Evaluation of Right-Turn Smart Channels Using Automated Traffic Conflict Analysis.\" Accident Analysis and Prevention (Accident Analysis and Prevention), no. 45 (2012): 120-130. Autey, Jarvis. Before and after traffic safety evaluations using computer vision techniques. Master Thesis. University of British Columbia (Vancouver), 2012. Baguley, C. J. \"The British traffic conflict technique: state of the art report.\" Leidschendam, The Netherlands, 1982. Proceedings, Third International Workshop on Traffic Conflicts Techniques. Institute for Road Safety Research. 107 Baker, William T. \"An evaluation of the traffic conflicts technique.\" Highway Research Record, 1972. Bowman, Bradley, Castiglone, and Yoder. Making advanced travel forecasting models affordable through model transferability. Final Report. Washington, DC: Prepared for Federal Highway Administration, US Department of Transportation, 2013. Brown, Gerald R. \"Traffic conflicts for road user safety studies.\" Canadian Journal of Civil Engineering 1, no. 21 (1994): 1-15. Cafiso, Salvatore, A. Garcia, Rosario Cavarra, and Mario Alfonso Romero Rojas. \"Pedestrian crossing safety improvements: before and after study using traffic conflict techniques.\" 4th International Symposium on Highway Geometric Design. Valencia, Spain, 2010. Cavallaro, Andrea, Olivier Steiger, and Touradj Ebrahimi. \"Tracking video objects in cluttered background.\" Circuits and Systems for Video Technology, IEEE Transactions on 15, no. 4, 2005: 575-584. Chin, Hoong-Chor, and Ser-Tong Quek. \"Measurement of traffic conflicts.\" Safety Science 3, no. 26 (1997): 169-185. Cooper, P. J. \"Experience with traffic conflicts in Canada with emphasis on “post encroachment time” techniques.\" International Calibration Study of Traffic Conflict Techniques. Springer Berlin Heidelberg, 1984. 75-96. Cunto, Flávio José Craveiro. Assessing safety performance of transportation systems using microscopic simulation. PhD dissertation, University of Waterloo, 2008. Cunto, Flávio, and Frank F. Saccomanno. \"Calibration and validation of simulated vehicle safety performance at signalized intersections.\" Accident Analysis & Prevention 3, no. 40 (2008): 1171-1179. 108 Davis, Gary A. \"Possible aggregation biases in road safety research and a mechanism approach to accident modeling.\" Accident Analysis & Prevention 6, no. 36 (2004): 1119-1127. de Leur, Paul, and Tarek Sayed. \"A framework to proactively consider road safety within the road planning process.\" Canadian Journal of Civil Engineering 4, no. 30 (2003): 711-719. de Leur, Paul, and Tarek Sayed. \"Using claims prediction model for road safety evaluation.\" Canadian Journal of Civil Engineering 5, no. 28 (2001): 804-812. Debnath, Ashim Kumar, and Hoong Chor Chin. \"Analysis of marine conflicts.\" the 19th KKCNN Symposium on Civil Engineering. 2006. Dijkstra, Atze, Paula Marchesini, and Frits Bijleveld. \"Do calculated conflicts in microsimulation model predict number of crashes?\" Transportation Research Record 1, no. 2147 (2010): 105-112. Duong, David, Frank Saccomanno, and Bruce Hellinga. \"Calibration of microscopic traffic model for simulating safety performance.\" 89th Annual Meeting of the Transportation Research Board. Washington, DC, 2010. El-Basyouny, K., and Tarek Sayed. \"Safety Performance Functions using Traffic Conflicts.\" Safety Science, no. 51 (2013): 160–164. Fan, Rong, Hao Yu, Pan Liu, and Wei Wang. \"Using VISSIM simulation model and Surrogate Safety Assessment Model for estimating field measured traffic conflicts at freeway merge areas.\" Intelligent Transport Systems. IET 7 1 (2013): 68-77. Fritzsche, Hans-Thomas, and Daimler-benz Ag. \"Amodel for traffic simulation.\" 21, no. 317 (1994). 109 Gettman, Douglas, and Larry Head. \"Surrogate safety measures from traffic simulation models.\" Transportation Research Record, no. 1840 (2003): 104-115. Gettman, Douglas, Lili Pu, Tarek Sayed, and Steven. Surrogate safety assessment model and validation: Final Report. FHWA-HRT-08-051, 2008. Glauz, William D., and Donald J. Migletz. Application of traffic conflict analysis at intersections. No. HS-028 882, 1980. Glennon, J. C., W. D. Glauz, M. C. Sharp, and B. A Thorson. \"Critique of the traffic-conflict technique.\" Transportation research record, no. 630 (1977): 32-38. Goldberg, D. E. . \"Genetic Algorithms in Search, Optimization, and Machine Learning.\" Addison-Wesley , Reading, Mass., 1989. Grayson, G.B., C. Hyden, J.H. Kraay, N. Muhlrad, and S. Oppe. The Malmö study: a calibration of traffic conflict techniques. Leidschendam: Institute for Road Safety Research, 1984. Gupte, Surendra, Osama Masoud, Robert FK Martin, and Nikolaos P. Papanikolopoulos. \"Detection and classification of vehicles.\" Intelligent Transportation Systems, IEEE Transactions on 3, no. 1, 2002: 37-47. Guttinger, V. A. \"Conflict observation techniques in traffic situations.\" Proceedings, First Workshop on Trafic Conflicts. 1977. Hauer, Ezra. \"The frequency–severity indeterminacy.\" Accident Analysis & Prevention 1, no. 38 (2006): 78-83. Hayward, John C. \"Near-miss determination through use of a scale of danger.\" Highway Research Record, no. 384 (1972). \"Highway Capacity Manual. ISBN 0-309-06681-6.\" Washington, D.C.: Transportation Research Board, 2000. 110 Hirst, W.M., L.J. Mountain, and M.J. Maher. \"Sources of error in road safety scheme evaluation: a method to deal with outdated accident prediction models.\" Accident Analysis & Prevention 5, no. 36 (2004): 717-727. Huang, Fei, Pan Liu, Hao Yu, and Wei Wang. \"Identifying if VISSIM simulation model and SSAM provide reasonable estimates for field measured traffic conflicts at signalized intersections.\" Accident Analysis & Prevention, no. 50 (2013): 1014-1024. Hummer, Joseph E. \"Traffic accident studies.\" In MANUAL OF TRANSPORTATION ENGINEERING STUDIES. Prentice-Hall, 1994. Hupfer, C. Computer aided image processing to modify traffic conflicts technique. Transportation Department, University of Kaiserslautern. Germany, 1997. Hydén, C., P. Gårder, and L. Linderholm. \"AN UPDATING OF THE USE AND FURTHER DEVELOPMENT OF THE TRAFFIC CONFLICTS TECHNIQUES.\" Proceedings of the Third International Workshop on Traffic Conflicts Techniques. 1982. Hydén, Christer. \"The development of a method for traffic safety evaluation: The Swedish Traffic Conflicts Technique.\" BULLETIN LUND INSTITUTE OF TECHNOLOGY, DEPARTMENT 70, 1987. Ismail, K., T. Sayed, and N. Saunier. \"A Methodology for Precise Camera Calibration for Data Collection Applications in Urban Traffic Scenes.\" Canadian Journal of Civil Engineering 1, no. 40 (2013): 57-67. Ismail, Karim. Application of computer vision techniques for automated road safety analysis and traffic data collection. Doctoral dissertation, UNIVERSITY OF BRITISH COLUMBIA (Vancouver), 2010. 111 Ismail, Karim, Tarek Sayed, and Nicolas Saunier. \"Automated Analysis of Pedestrian-Vehicle: Conflicts Context for Before-and-After Studies.\" Transportation Research Record: Journal of the Transportation Research Board, no. 2198 (2010): 52-64. Laureshyn, Aliaksei, Håkan Ardö, Åse Svensson, and Thomas Jonsson. \"Application of automated video analysis for behavioural studies: concept and experience.\" IET Intelligent Transport Systems 3, no. 3, 2009: 345-357. Lou, Jianguang, Tieniu Tan, Weiming Hu, Hao Yang, and Steven J. Maybank. \"3-D model-based vehicle tracking.\" Image Processing, IEEE Transactions on 14, no. 10, 2005: 1561-1569. Malaterre, G., and N. Muhlrad. \"International comparative study on traffic conflict techniques.\" In Proceedings, Second International Trajjic Conflict Technique Workshop. 1979. 185-231. Malinovskiy, Yegor, Jianyang Zheng, and Yinhai Wan. \"A simple and model-free algorithm for real-time pedestrian detection and tracking.\" In 86th Annual Meeting of the Transportation Research Board. Washington, DC., 2007. Minderhoud, Michiel M., and Piet HL Bovy. \"Extended time-to-collision measures for road traffic safety assessment.\" Accident Analysis & Prevention 1, no. 33 (2001): 89-97. Muhlrad, N. \"The French traffic conflict technique: a state of the art report.\" Muhlrad, N. \"The French traffic conflict technique: a state of the art report.\" JH 32 Kraay, Proceedings of the Third International Workshop on Traffic Conflicts Techniques. The. Vol. 33. 1982. 1982. Nicholson, A. J. \"The variability of accident counts.\" Accident Analysis & Prevention 1, no. 17 (1985): 47-56. 112 Oppe, S. \"What task is a traffic conflict technique intended for?\" Proceedings of 1st Workshop on Traffic Conflicts. Olso, Norway, 1977. Panwai, Sakda, and Hussein Dia. \"Comparative evaluation of microscopic car-following behavior.\" Intelligent Transportation Systems, no. IEEE Transactions on 6, no. 3 (2005): 314-325. Park, Byungkyu Brian, and Jongsun Won. . Microscopic simulation model calibration and validation handbook. No. FHWA/VTRC 07-CR6, 2006. Park, Byungkyu, and Hongtu Qi. Development and evaluation of a calibration and validation procedure for microscopic simulation models. No. FHWA/VTRC 05-CR1, 2004. Park, Byungkyu, and Hongtu Qi. . \"Development and Evaluation of a Procedure for the Calibration of Simulation Models.\" Transportation Research Record, no. 1934 (2005): 208-217. Park, Byungkyu, Jongsun Won, and Ilsoo Yun. \"Application of microscopic simulation model Calibration and validation procedure: Case study of coordinated actuated signal system.\" Transportation Research Record, no. 1978 (2006): 113-122. Perkins, Stuart R., and Joseph L. Harris. \"Traffic conflict characteristics-accident potential at intersections.\" Highway Research Record 225, 1968. \"Quadstone Paramics User Guide - 6.9.3.\" 2013. \"Quadstone PARAMICS User Manual - V.6.9.3 - Paramics Product Suite>Analyser.\" 2013. \"Quadstone PARAMICS User Manual - V.6.9.3 - Paramics Product Suite>Modeller.\" 2013. \"Quadstone PARAMICS User Manual - V.6.9.3 - Paramics Product Suite>Modeller>Technical Notes>Actuated signal Controller.\" 2013. \"Quadstone PARAMICS User Manual - V.6.9.3.\" 2013. 113 Sacchi, E., Tarek Sayed, and P. DeLeur. \"A Comparison of Collision-Based and Conflict-Based Safety Evaluations: The Case of Right-Turn Smart Channels.\" Accident Analysis and Prevention, no. 59 (2013): 260-266. Saleem, Taha, Bhagwant Persaud, Amer Shalaby, and Alexander Ariz. \"Can Microsimulation Be Used to Estimate Intersection Safety? Case Studies Using VISSIM and Paramics.\" Transportation Research Record: Journal of the Transportation Research Board, no. 2432 (2014): 142-148. Saunier, N., and Tarek Sayed. \"Automated Analysis of Road Safety with Video Data.\" Transportation Research Record: Journal of the Transportation Research Board, no. 2019 (2007): 57-64. Saunier, N., and Tarek Sayed. \"Probabilistic Framework for the Automated Analysis of the Exposure to Road Collision.\" Transportation Research Record: Journal of the Transportation Research Board, no. 2083 (2008): 96–104. Saunier, N., Tarek Sayed, and K. Ismail. \"Large Scale Automated Analysis of Vehicle Interactions and Collisions.\" Transportation Research Record: Journal of the Transportation Research Board, no. 2147 (2010): 42-50. Saunier, Nicolas, and Tarek Sayed. \"A feature-based tracking algorithm for vehicles in intersections.\" Computer and Robot Vision, 2006. The 3rd Canadian Conference on. IEEE. 2006. Saunier, Nicolas, Tarek Sayed, and Clark Lim. . \"Probabilistic collision prediction for vision-based automated road safety analysis.\" Intelligent Transportation Systems Conference, ITSC 2007. IEEE, 2007: 872-878. 114 Sawalha, Ziad, and Tarek Sayed. \"Transferability of accident prediction models.\" Safety science 3, no. 44 (2006): 209-219. Sayed, Tarek, and S. Zein. \"Traffic Conflict Standards for Intersections.\" Transportation Planning and Technology (Transportation Planning and Technology) 22 (1999): 309-323. Sayed, Tarek, G. Brown, and F. Navin. \"Simulation of Traffic Conflicts at Unsignalized Intersections With TSC-Sim.\" Accident Analysis and Prevention (Accident Analysis and Prevention) 5, no. 26 (1994): 593-607. Sayed, Tarek, K. Ismail, M. Zaki, and J. Autey. \"Feasibility of Computer Vision-Based Safety Evaluations: Case Study of a Signalized Right-Turn Safety Treatment.\" Journal of the Transportation Research Board”, Transportation Research Record, no. 2280 (2012). Sayed, Tarek, M. Zaki, and J. Autey. \"Automated Safety Diagnosis of Vehicle-Bicycle Interactions using Computer Vision Analysis.\" Safety Science, no. 59 (2013): 163-172. Shahdah, Usama, Frank Saccomanno, and Persaud Bhagwant . \"Integrated traffic conflict model for estimating crash modification factors.\" Accident Analysis & Prevention, no. 71 (2014): 228-235. Sikder, Sujan, Bertho Augustin, Abdul Rawoof Pinjari, and Naveen Eluru. \"Spatial Transferability of Tour-Based Time-of-Day Choice Models.\" Transportation Research Record: Journal of the Transportation Research Board 1, no. 2429 (2014): 99-109. So, Jaehyun Jason, George Dedes, Byungkyu Brian Park, Siavash HosseinyAlamdary, and Dorota Grejner-Brzezinsk. \"Development and evaluation of an enhanced surrogate safety assessment framework.\" Transportation research part C: emerging technologies, no. 50 (2015): 51-67. 115 Sobhani, Amir, William Young, and Majid Sarvi. \"A simulation based approach to assess the safety performance of road locations.\" Transportation Research Part C: Emerging Technologies, no. 32 (2013): 144-158. Songchitruksa, Praprut, and Andrew P. Tarko. \"The extreme value theory approach to safety estimation.\" Accident Analysis & Prevention 4, no. 38 (2006): 811-822. Stevanovic, Aleksandar, Jelka Stevanovic, Jaehyun So, and Marija Ostojic. \"Multi-criteria optimization of traffic signals: Mobility, safety, and environment.\" Transportation Research Part C: Emerging Technologies, no. 55 (2015): 46-68. Svensson, Ase. A method for analysing the traffic process in a safety perspective. PhD Thesis, Lund University, 1998. Svensson, Ase, and Hydén Christer. \"Estimating the severity of safety related behaviour.\" Accident Analysis & Prevention 2, no. 38 (2006): 379-385. Tageldin, Ahmed, Tarek Sayed, Mohamed Zaki, and Mohamed Azab. \"A Safety Evaluation of an Adaptive Traffic Signal Control System using Computer Vision.\" Advances in Transportation Studies 2, no. Special Issue (2014): 83-96. Tarko, Andrew P., and Praprut Songchitruksa. \"Estimating frequency of crashes as extreme traffic events.\" Annual Meeting of the Transportation Research Board. Washington, DC., 2005. Tarko, Andrew, Gary Davis, Nicolas Saunier, Tarek Sayed, and Simon Washington. \"Surrogate measures of safety.\" White paper, ANB20 (3) Subcommittee on Surrogate Measures of Safety, 2009. 116 Transport Canada. \"Canadian Motor Vehicle Traffic Collision Statistics: 2013.\" Motor Vehicle Safety. 2013. http://www.tc.gc.ca/media/documents/roadsafety/cmvtcs2013_eng.pdf (accessed July 10, 2015). —. Road Safety in Canada. 2011. http://www.tc.gc.ca/eng/motorvehiclesafety/tp-tp15145-1201.htm (accessed July 10, 2015). van der Horst, A.R.A., and Gerald R. Brown. TIME-TO-COLLISION AND DRIVER DECISION MAKING IN BRAKING. Technical report, TNO Institute for Perception, 1989. Van der Horst, Adrianus Rigardus Antonius. A time-based analysis of road user behaviour in normal and critical encounters. PhD Thesis. TU Delft, Delft University of Technology, 1990. van der Horst, Richard, and Jeroen Hogema. \"Time-to-collision and collision avoidance systems.\" In Proceedings of the 6th ICTCT Workshop. 1993. VISSIM 5.30-05 User Manual. Karlsruhe, Germany: PTV Planung Transport Verkehr AG, 2011. VISSIM 5.40-01 COM Interface Manual. AG: Planung Transport Verkehr , 2011. Wang, Chen, and Nikiforos Stamatiadis. . \"Evaluation of a simulation-based surrogate safety metric.\" Accident Analysis & Prevention, no. 71 (2014): 82-92. Wiedemann, R. \"Modeling of RTI-Elements on multi-lane roads.\" Advanced Telematics in Road Transport edited by the Commission of the European Community, DG XIII, Brussels , 1991. Wiedemann, R. \"Simulation des Straßenverkehrsflusses.\" . Schriftenreihe des Instituts für Verkehrswesen der Universität Karlsruhe, Heft 8. , 1974. Williams, Martin J. \"Validity of the traffic conflicts technique.\" Accident Analysis & Prevention 2, no. 13 (1981): 133-145. 117 World Health Organization. Road safety: a public health issue. March 29, 2004. http://www.who.int/features/2004/road_safety/en/ (accessed July 10, 2015). Zaki, Mohamed H., and Tarek Sayed. \"Automated Classification of Road-User Movement Trajectories.\" CSCE International Transportation Specialty Conference. Edmonton, AB, 2012. Zaki, Mohamed, Tarek Sayed, Ahmed Tageldin, and Mohamed Hussein. \"Application of Computer Vision to Diagnosis of Pedestrian Safety.\" Transportation Research Record: Journal of the Transportation Research Board, no. 2393 (2013): 75-84. 118 Appendices Appendix A Traffic Data Details This appendix contains traffic data details for the first and the second intersections. Traffic volumes, platoon ratios, delays, and signal programs are provided in the following sections. A.1 Traffic Volumes Hour Passenger Car Total Left Through Right 1 120 413 52 585 2 123 482 61 666 3 207 703 89 999 4 177 868 110 1155 5 195 836 106 1137 6 237 985 125 1347 7 204 985 125 1314 8 204 916 116 1236 9 213 974 124 1311 Table A.1 Traffic volume of the first intersection (west approach) Hour Passenger Car Total Left Through Right 1 108 662 76 846 2 75 619 71 765 3 138 686 79 903 4 78 759 87 924 5 84 853 98 1035 6 105 691 80 876 7 117 872 100 1089 8 66 648 75 789 9 90 823 95 1008 Table A.2 Traffic volume of the first intersection (east approach) 119 Hour Passenger Car Total Left Through Right 1 72 316 98 486 2 84 334 104 522 3 105 412 128 645 4 99 405 126 630 5 129 396 123 648 6 216 524 163 903 7 141 618 192 951 8 144 598 185 927 9 168 641 199 1008 Table A.3 Traffic volume of the first intersection (north approach) Hour Passenger Car Total Left Through Right 1 90 304 56 450 2 87 256 47 390 3 111 406 74 591 4 129 362 67 558 5 129 337 62 528 6 117 421 77 615 7 186 476 88 750 8 168 535 98 801 9 174 484 89 747 Table A.4 Traffic volume of the first intersection (south approach) 120 Hour Passenger Car Total Left Through Right 1 76 556 129 761 2 67 472 126 665 3 86 635 137 858 4 89 650 151 890 5 99 730 158 987 6 108 876 97 1081 7 117 938 117 1172 8 115 919 115 1149 9 110 883 110 1103 Table A.5 Traffic volume of the second intersection (west approach) Hour Passenger Car Total Left Through Right 1 83 693 50 826 2 76 634 45 755 3 87 722 61 870 4 92 759 64 915 5 91 748 73 912 6 95 747 104 946 7 107 873 85 1065 8 105 806 136 1047 9 106 826 127 1059 Table A.6 Traffic volume of the second intersection (east approach) 121 Hour Passenger Car Total Left Through Right 1 35 217 93 345 2 38 248 90 376 3 42 265 113 420 4 47 299 121 467 5 46 306 111 463 6 57 381 131 569 7 52 334 136 522 8 68 484 129 681 9 74 534 133 741 Table A.7 Traffic volume of the second intersection (north approach) Hour Passenger Car Total Left Through Right 1 40 311 48 399 2 35 259 52 346 3 35 268 45 348 4 32 262 29 323 5 38 289 53 380 6 40 312 43 395 7 48 352 82 482 8 43 326 60 429 9 45 342 63 450 Table A.8 Traffic volume of the second intersection (south approach) 122 A.2 Platoon Ratios and Delays West Approach East Approach Hour % of Vehicles on green time Platoon Ratio Average Delay (s) Hour % of Vehicles on green time Platoon Ratio Average Delay (s) 1 0.55 1.18 11.9 1 0.57 1.25 15.8 2 0.67 1.34 10 2 0.68 1.46 11 3 0.33 0.7 20.5 3 0.28 0.62 23.9 4 0.5 1.04 12 4 0.37 0.84 22.96 5 0.59 1.15 12.31 5 0.36 0.74 26.7 6 0.4 0.86 26.95 6 0.4 0.96 21.21 7 0.24 0.53 27.58 7 0.27 0.68 23.7 8 0.33 0.69 25.42 8 0.31 0.71 23.01 9 0.41 0.88 28.37 9 0.26 0.61 29.24 Table A.9 Platoon ratios and delays of the first intersection (east and west approaches) North Approach South Approach Hour % of Vehicles on green time Platoon Ratio Average Delay (s) Hour % of Vehicles on green time Platoon Ratio Average Delay (s) 1 0.39 1.14 18.76 1 0.39 1.14 18.61 2 0.33 0.94 20.91 2 0.41 1.19 20.72 3 0.45 1.29 20.97 3 0.38 1.09 20.3 4 0.57 1.6 13.87 4 0.41 1.15 18.9 5 0.57 1.7 14.28 5 0.36 1.06 21.21 6 0.37 0.99 20.69 6 0.24 0.66 22.44 7 0.48 1.31 19.74 7 0.39 1.07 19.49 8 0.47 1.3 20.55 8 0.39 1.08 20.12 9 0.43 1.21 22.02 9 0.41 1.15 19.21 Table A.10 Platoon ratios and delays of the first intersection (north and south approaches) 123 West Approach East Approach Hour % of Vehicles on green time Platoon Ratio Average Delay (s) Hour % of Vehicles on green time Platoon Ratio Average Delay (s) 1 0.55 1.04 10.54 1 0.55 1.03 10.95 2 0.68 1.23 4.73 2 0.54 0.97 10.63 3 0.75 1.39 4.33 3 0.34 0.63 21.77 4 0.78 1.5 4.06 4 0.33 0.63 22.01 5 0.78 1.47 4.33 5 0.45 0.86 14.37 6 0.69 1.4 6.8 6 0.39 0.78 20.96 7 0.66 1.37 7.73 7 0.58 1.21 9.34 8 0.67 1.46 8.24 8 0.63 1.37 9.02 9 0.62 1.38 9.58 9 0.44 0.98 16.69 Table A.11 Platoon ratios and delays of the second intersection (east and west approaches) 124 A.3 Traffic Signal Programs Hour Cycle Length (Sec) 72nd Avenue 128th Street Max Green Time (Through) Min Green Time (Through) (sec) Protected Left Turn (sec) Max Green Time (Through) Min Green Time (Through) (sec) Protected Left Turn (sec) 1 90 44 33 7 34 23 7 2 90 44 33 7 34 23 7 3 95 47 35 8 36 25 7 4 95 46 35 7 37 26 7 5 95 48 37 7 35 24 7 6 95 45 34 7 38 27 7 7 95 45 34 7 38 27 7 8 95 46 35 7 37 26 7 9 95 46 35 7 37 26 7 Table A.12 Traffic signal program of the first intersection Hour Cycle Length (Sec) 72nd Avenue 132nd Street Max Green Time (Through) Min Green Time (Through) (sec) Protected Left Turn (sec) Max Green Time (Through) Min Green Time (Through) (sec) Protected Left Turn (sec) 1 90 59 44 11 19 8 7 2 90 58 44 10 20 9 7 3 95 56 45 7 27 16 7 4 92 59 45 10 21 10 7 5 95 56 45 7 27 16 7 6 95 55 42 9 28 17 7 7 95 54 41 9 29 18 7 8 92 47 36 7 33 22 7 9 95 54 41 9 29 18 7 Table A.13 Traffic signal program of the second intersection 125 Appendix B Field-Measured Conflict Data This appendix contains field-measured rear-end conflicts estimated from the automated video based computer vision technique. The following tables provide details of these conflicts for the first and the second intersections. Table B.1 Field-measured traffic conflicts of the first intersection (calibration dataset) Section-HourTTC≤1TTC≤1.1TTC≤1.2TTC≤1.3TTC≤1.4TTC≤1.5TTC≤1.6TTC≤1.7TTC≤1.8TTC≤1.9TTC≤2TTC≤2.1TTC≤2.2TTC≤2.3TTC≤2.4TTC≤2.5TTCv2.6TTC≤2.7TTC≤2.8TTC≤2.9TTC≤31 5 5 5 6 6 6 7 7 8 8 10 11 13 13 13 15 18 19 20 20 202 2 2 3 3 3 4 4 4 4 7 8 8 9 11 14 16 17 18 18 20 213 9 9 9 10 11 11 12 12 12 12 14 15 16 16 16 17 17 17 18 20 224 5 5 8 10 10 10 11 12 12 13 15 18 21 25 26 27 29 30 31 35 415 6 6 8 8 9 9 9 9 9 9 11 11 11 12 12 12 12 12 12 12 126 2 3 4 4 4 4 5 5 5 5 6 6 8 8 8 10 11 11 13 16 177 5 5 5 7 7 8 9 9 10 10 10 12 14 15 16 17 17 19 21 21 238 2 2 2 2 3 3 4 4 4 4 5 6 6 7 8 9 9 9 11 11 129 3 3 3 3 3 3 3 4 4 4 5 5 7 9 10 12 14 14 15 15 1510 10 10 11 11 12 13 13 14 14 15 18 19 20 22 24 25 28 31 33 35 3611 8 8 9 10 11 13 15 16 17 18 20 21 23 24 25 26 27 31 32 32 3212 3 3 3 3 3 3 3 6 6 7 7 8 8 9 10 11 14 17 18 18 1913 9 11 12 13 15 16 18 19 20 21 23 23 23 26 29 29 30 30 32 34 3414 16 16 17 19 19 20 21 21 23 24 24 25 25 25 26 26 27 31 33 34 3415 29 29 30 32 34 35 40 41 43 46 51 54 58 58 62 64 66 70 73 78 8116 19 22 23 23 26 27 27 30 32 33 35 39 42 42 45 47 51 55 58 60 6217 20 22 23 26 26 29 30 32 36 37 40 40 40 41 41 43 43 44 44 44 4518 12 12 13 16 17 17 19 21 24 25 26 26 27 31 39 41 44 51 54 57 5919 19 20 21 23 26 26 27 29 31 35 38 40 43 48 54 60 67 72 73 79 8720 13 15 16 16 16 17 19 20 23 23 24 24 25 26 28 29 30 30 30 36 3821 13 14 16 17 18 21 21 23 27 27 30 31 32 35 39 43 46 49 54 56 5822 3 3 3 3 3 3 4 4 5 5 5 5 5 5 6 6 6 6 7 8 823 2 2 2 2 2 2 2 2 2 2 2 2 2 3 4 4 4 4 4 4 424 5 5 5 5 5 5 5 5 5 6 6 6 7 7 7 7 7 8 8 9 1025 1 1 1 1 1 1 1 2 3 3 3 3 3 3 3 3 3 3 3 3 326 0 0 0 1 1 1 1 1 1 1 1 3 5 6 6 6 7 7 9 9 927 2 2 2 2 2 2 2 2 2 2 2 2 4 4 4 7 7 8 10 10 1228 7 8 11 11 12 12 14 17 20 22 23 25 26 28 30 31 33 35 36 37 3829 1 1 1 1 1 1 1 1 2 4 4 5 6 9 10 10 10 10 12 12 1230 2 2 2 2 3 3 5 5 5 6 6 6 8 8 8 10 11 11 12 12 12126 Table B.2 Field-measured traffic conflicts of the first intersection (validation dataset) Section-HourTTC≤1TTC≤1.1TTC≤1.2TTC≤1.3TTC≤1.4TTC≤1.5TTC≤1.6TTC≤1.7TTC≤1.8TTC≤1.9TTC≤2TTC≤2.1TTC≤2.2TTC≤2.3TTC≤2.4TTC≤2.5TTCv2.6TTC≤2.7TTC≤2.8TTC≤2.9TTC≤31 5 6 9 10 10 11 13 15 15 19 20 22 24 26 27 28 28 29 29 31 322 3 4 4 4 5 7 10 11 15 17 17 19 20 21 21 22 22 23 25 26 273 5 5 5 6 10 10 13 15 15 16 17 18 20 20 22 23 25 25 27 29 314 14 15 16 19 20 20 21 25 30 31 33 33 34 36 39 40 41 41 41 43 445 3 3 3 3 4 4 5 5 6 6 10 10 13 13 15 17 19 21 22 22 226 7 11 12 14 16 22 23 25 26 27 30 32 33 37 41 41 45 46 48 49 507 2 2 2 2 3 4 4 5 5 6 8 8 8 8 8 10 11 11 12 12 138 1 1 1 1 2 2 2 2 2 4 4 5 5 5 5 6 8 8 8 8 89 4 4 4 5 5 5 5 5 5 8 8 8 10 11 12 12 12 14 16 16 1710 1 1 1 1 2 2 2 3 5 6 7 8 9 11 13 13 13 14 15 15 1711 2 2 2 2 2 2 2 2 3 3 5 6 7 9 13 14 15 18 18 21 2112 2 3 3 3 3 3 4 5 5 6 6 8 8 9 10 12 13 13 14 14 1513 4 4 4 4 4 4 4 5 6 6 8 9 11 13 15 17 18 22 22 23 2414 2 2 2 2 2 2 2 2 2 3 4 5 6 7 7 8 9 9 11 12 1215 4 4 4 4 4 4 4 4 5 6 6 8 9 10 10 10 11 13 16 20 2016 2 5 7 9 10 11 12 12 13 13 13 14 15 20 23 25 26 28 30 32 3317 1 2 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 6 7 7 818 0 0 1 2 2 2 4 5 5 5 5 7 8 8 8 8 8 8 8 9 1119 0 0 1 2 2 4 5 6 7 7 7 8 9 11 12 14 17 17 17 17 1720 1 1 3 3 4 4 4 6 7 8 8 8 9 11 12 12 14 14 14 14 1421 18 19 21 23 25 27 31 35 37 39 39 41 41 45 46 48 50 50 54 57 6022 12 14 15 20 24 30 33 38 40 44 50 51 54 57 63 65 65 70 72 75 7823 14 17 20 22 25 28 31 33 38 41 41 46 53 56 62 63 64 66 68 71 7224 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 025 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 126 0 1 1 1 1 1 1 1 1 1 1 1 2 2 2 3 4 4 4 5 527 0 0 2 2 2 2 2 2 2 3 3 3 3 4 4 4 4 4 6 6 628 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 4 5 5 5 5 529 1 1 1 1 1 1 1 2 2 2 4 5 6 6 7 8 11 12 12 12 1230 2 2 2 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 5127 Table B.3 Field-measured traffic conflicts of the second intersection Section-HourTTC≤1TTC≤1.1TTC≤1.2TTC≤1.3TTC≤1.4TTC≤1.5TTC≤1.6TTC≤1.7TTC≤1.8TTC≤1.9TTC≤2TTC≤2.1TTC≤2.2TTC≤2.3TTC≤2.4TTC≤2.5TTCv2.6TTC≤2.7TTC≤2.8TTC≤2.9TTC≤31 6 6 6 6 6 8 9 9 10 10 10 10 10 12 12 12 13 15 15 16 162 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 23 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 7 7 7 7 74 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 4 5 5 5 55 1 1 1 1 1 1 1 2 2 3 4 4 4 5 7 7 7 8 8 8 86 5 5 5 5 5 5 9 9 9 9 9 9 9 10 11 11 12 12 12 12 137 5 5 6 6 6 6 6 6 6 7 7 8 8 9 9 9 10 10 10 11 118 0 0 0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 29 3 3 4 4 4 4 4 5 5 5 5 5 5 5 5 5 5 5 5 5 510 1 1 1 3 3 3 3 3 3 3 3 4 4 4 4 4 4 5 5 5 511 2 2 2 2 2 2 2 3 4 4 4 4 4 4 4 5 5 5 5 5 512 3 3 3 3 3 5 5 5 5 5 7 8 9 10 11 11 12 12 12 12 1213 5 8 10 14 15 16 17 17 18 18 21 21 21 22 22 25 26 27 27 28 2914 0 0 0 0 0 0 1 2 3 3 4 4 4 6 6 6 6 6 7 7 715 7 9 9 10 11 12 12 12 13 13 14 17 19 21 21 22 24 24 25 26 2716 4 4 4 4 4 5 5 6 8 10 13 13 14 14 14 15 17 18 19 19 1917 3 3 3 3 5 5 5 5 7 7 7 7 7 8 8 8 10 14 15 15 1518 5 6 6 7 7 9 10 10 13 15 16 17 18 19 20 22 27 30 31 31 3219 2 3 3 4 6 8 10 12 13 17 21 21 22 24 26 28 30 32 35 36 3920 28 28 29 31 31 31 32 32 33 33 36 39 40 42 45 48 49 49 50 51 5321 18 18 20 23 25 25 25 28 30 31 32 34 34 35 37 37 37 41 43 44 4422 23 28 29 30 31 36 37 40 44 51 52 58 59 62 65 70 75 75 76 77 7723 17 18 18 18 19 20 20 21 21 22 23 23 26 27 28 29 30 31 32 33 35128 Appendix C Sensitivity Analysis This appendix contains the details of the sensitivity analysis of the VISSIM parameters. Sensitivity analysis was done on 31 parameters to identify the parameters which have the biggest effect on the simulated rear-end conflicts. Table C.1 Sensitivity analysis of the car following model parameters Parameter Unit Default Values 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 80.5 7 10 11 4 10 18 29 29 44 41 51 25 63 59 138 109 285 334 305 298 463 430 803 6791 3 1 6 3 8 12 12 10 18 30 27 20 44 26 72 79 279 314 260 291 427 411 701 6281.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5952 2 5 9 2 8 6 22 17 11 17 14 20 37 28 65 65 251 287 261 264 369 345 522 5982.5 5 3 4 4 13 11 12 10 19 15 16 14 37 33 48 74 258 264 245 222 359 361 495 5590.5 15 11 10 10 26 20 33 39 34 39 43 34 80 57 91 117 230 255 232 204 349 315 437 5320.7 3 9 5 6 13 5 16 20 21 21 24 22 35 41 55 80 235 279 268 261 347 361 513 5960.9 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5951.1 2 6 2 2 3 7 12 9 14 21 11 14 26 26 61 63 311 365 295 310 447 443 710 6581.3 3 1 3 1 7 3 11 9 11 13 23 10 35 25 45 44 332 385 323 342 530 491 688 7562 0 0 4 2 4 4 7 7 11 9 17 13 28 27 63 51 250 289 253 221 410 409 642 5403 3 4 5 1 8 4 13 13 12 20 21 10 41 26 61 65 281 308 269 232 395 400 609 5914 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5955 6 2 4 2 9 8 21 16 16 18 17 15 34 34 66 74 253 324 265 266 391 397 594 6136 2 3 7 1 15 6 11 10 14 16 19 16 36 21 52 61 285 302 262 287 380 399 592 625-4 5 3 10 1 8 11 6 14 21 15 30 13 37 36 69 60 280 308 267 273 393 354 587 631-6 4 3 6 2 6 9 18 18 18 23 25 21 24 30 64 83 249 312 254 222 352 349 551 588-8 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-10 3 10 5 4 2 6 16 15 13 28 16 17 11 26 55 69 270 297 263 247 342 396 603 581-12 4 0 3 0 4 6 10 6 14 15 26 18 25 26 49 58 271 281 283 243 375 394 543 557(-/+ 0.01) 2 8 4 0 13 6 13 10 13 23 22 6 27 24 55 60 248 291 245 230 386 388 558 638(-/+ 0.1) 6 5 7 4 6 9 14 14 15 21 21 16 19 35 53 63 277 331 262 262 353 384 588 594(-/+ 0.35) 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595(-/+ 1) 4 0 2 0 4 2 2 13 25 25 23 17 43 36 64 77 283 329 270 298 386 452 590 649(-/+ 1.5) 2 2 2 0 0 4 11 4 15 19 16 15 53 30 89 83 276 323 296 251 469 397 661 6844 7 1 13 3 11 4 13 11 24 12 25 14 36 25 62 51 277 321 277 250 411 409 579 5878 7 6 4 3 14 11 20 13 20 24 14 13 47 39 81 68 290 285 262 252 440 399 598 63311.44 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59516 1 8 9 4 16 7 14 12 10 24 23 14 44 23 63 56 269 305 251 280 415 401 627 58120 9 10 6 3 14 15 18 14 26 28 23 10 41 38 59 67 274 334 270 270 419 446 588 6310.15 2 5 1 2 6 6 23 16 19 19 14 13 32 38 67 66 270 347 284 250 390 384 657 6210.2 2 0 8 3 5 7 14 13 12 8 27 15 28 38 52 60 263 274 298 241 374 370 601 6130.25 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5950.3 4 4 5 4 16 6 11 7 17 24 24 15 43 25 45 48 274 315 293 258 412 398 538 6070.4 2 3 6 8 9 3 19 10 10 18 23 20 39 16 81 72 240 339 259 251 395 398 648 593123(-/+ 0.35)11.44#4567 0.25CC4&CC5m/s²CC6CC7TTC < 34-8CC0CC1CC2CC3msecmCar Following Model (Widemann 99)1.50.9TTC < 1 TTC < 1.5129 Table C.2 Sensitivity analysis of the car following model parameters (Continued) Parameter Unit Default Values 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 82.5 1 4 5 4 8 4 21 7 13 17 22 15 35 26 58 69 282 303 269 227 368 399 581 6003 6 1 2 1 7 12 14 17 29 12 14 18 30 30 62 74 289 296 254 247 379 392 622 5763.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5954 6 4 2 1 2 6 10 17 13 22 23 8 19 23 55 73 288 323 277 247 369 392 615 6424.5 3 3 2 1 7 6 18 9 15 14 25 11 27 32 56 65 275 298 274 236 413 402 590 5990.8 8 10 4 5 7 9 20 18 27 36 18 21 25 38 84 70 285 337 281 250 397 389 645 6221.15 5 6 5 0 12 11 9 22 19 20 25 19 35 43 49 74 290 329 285 277 394 416 606 6281.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5951.85 7 1 5 4 5 9 15 11 18 16 21 17 32 37 53 66 271 300 271 251 398 394 552 6202.2 4 4 6 1 6 10 17 18 12 19 21 11 28 37 48 82 281 303 277 235 350 396 593 7020 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59510 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59515 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59520 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59530 10 4 2 2 3 5 17 11 19 18 21 19 38 28 59 55 254 337 285 256 417 392 591 58850 3 1 0 7 6 8 10 11 15 17 13 26 34 27 52 80 302 337 271 238 390 353 563 661150 3 3 4 3 9 3 20 17 24 16 18 12 52 32 60 70 289 301 269 283 420 404 602 605250 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595300 6 5 3 4 7 7 17 6 20 14 20 11 36 30 59 65 292 309 274 269 377 406 559 612350 8 1 3 2 7 14 19 14 21 18 16 14 27 36 60 61 285 340 271 248 393 438 618 5492 3 5 6 1 11 7 12 16 14 14 21 21 41 28 64 57 278 340 289 273 408 409 602 6523 2 5 6 6 5 9 17 15 14 24 24 14 32 41 53 68 269 345 284 245 407 419 571 5824 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5955 2 3 3 3 8 6 12 14 13 17 12 15 21 26 62 74 279 317 262 252 347 371 606 6286 5 3 5 1 7 7 18 16 13 19 21 19 37 29 69 76 235 333 270 253 408 383 594 6270 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59510 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59515 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59520 10 4 2 2 4 5 12 10 19 18 21 19 38 28 65 55 254 337 286 258 416 391 601 59530 10 4 2 2 4 5 12 10 19 18 21 19 38 28 65 55 254 337 286 258 416 391 601 59550 7 0 4 7 11 6 19 19 18 13 19 26 30 32 62 61 272 321 274 272 373 411 615 590100 5 2 4 1 2 4 9 13 13 17 17 16 37 29 55 68 266 336 288 285 403 404 582 616150 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595200 10 4 2 2 3 7 14 11 19 18 21 19 38 23 62 61 254 337 286 258 404 389 626 593250 3 3 1 4 7 3 18 10 16 16 14 16 21 25 57 59 248 339 282 241 369 369 602 6110&0 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5950.5 & 0.05 4 5 2 2 4 10 11 13 12 15 19 12 20 32 59 79 245 307 267 260 380 403 653 6170.5 &0.10 6 4 0 3 11 5 14 13 15 19 21 20 41 30 57 64 278 279 269 256 415 395 576 6191&0.05 5 5 5 2 20 8 12 15 15 17 20 10 38 32 61 81 243 319 276 276 400 415 614 6221&0.10 6 4 1 2 7 2 25 16 16 17 22 20 24 18 70 64 295 285 271 260 394 406 651 613TTC < 3Car Following Model (Widemann 99) TTC < 1 TTC < 1.540mObserved Vehicles150Look back distance (max)mmLook back distance (min)m/s²m/s²CC8CC93.51.5Look ahead distance (min)0#891011Temporary lack of attention duration & percentagsec & %0 & 014151213250Look ahead distance (max)mm130 Table C.3 Sensitivity analysis of the lane change parameters Parameter Unit Default Values 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8-6 10 3 2 2 2 8 19 13 19 17 20 18 34 33 66 71 257 321 286 257 398 391 597 631-5 10 4 2 2 2 7 23 16 19 17 18 19 33 34 45 74 256 336 283 256 399 392 581 617-4 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-3 5 4 2 2 2 7 11 18 14 18 19 18 33 32 58 87 246 335 285 257 406 389 577 627-2 6 1 2 3 6 4 18 18 14 16 17 18 36 25 56 64 246 345 281 252 414 393 590 603-5 7 4 4 2 11 5 22 15 24 24 19 17 36 28 77 70 270 324 277 266 413 405 602 615-4 7 4 3 2 12 5 15 13 19 24 17 17 29 38 62 70 274 326 273 266 396 390 608 598-3 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-2 9 3 3 4 11 8 15 13 17 22 17 15 30 28 47 61 257 327 278 258 412 386 559 590-1 6 1 7 2 5 6 14 13 13 16 20 15 28 31 55 61 265 334 300 242 384 402 593 59350 10 3 2 2 4 5 16 16 19 19 21 19 38 28 58 76 255 323 286 258 416 393 587 63575 10 4 2 2 4 5 14 10 19 18 21 19 38 28 67 55 253 336 286 258 416 391 591 595100 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595125 10 4 2 2 4 5 12 10 19 18 21 19 38 28 63 56 254 337 286 258 417 391 587 604150 10 4 2 2 4 8 10 15 19 18 21 19 38 33 49 61 254 337 286 258 420 392 567 59250 10 4 2 2 4 5 14 10 19 18 21 19 38 28 69 58 254 337 286 258 416 391 590 60475 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 596100 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595125 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 597 596150 10 4 2 2 3 5 18 14 19 18 21 19 36 28 65 58 254 337 286 258 408 391 581 612-1.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-1.25 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-1 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-0.75 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-0.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-1.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-1.25 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-1 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-0.75 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-0.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59510 6 4 2 2 3 4 18 9 19 17 21 19 38 27 57 60 259 327 286 257 385 395 615 59535 10 4 2 2 4 4 11 9 19 18 21 19 36 27 58 53 254 337 286 258 404 394 609 59860 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 59585 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595110 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5950.3 9 3 8 6 3 6 7 22 20 17 22 15 32 30 66 82 252 341 271 265 374 394 616 6100.4 9 3 2 2 3 5 20 20 18 17 20 20 32 25 55 79 259 343 285 259 390 394 602 6570.5 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5950.75 4 4 2 3 4 2 14 22 13 18 18 18 41 20 51 68 258 330 296 250 425 393 608 6261 3 2 6 2 13 7 15 12 13 20 17 18 24 23 63 63 258 346 276 260 395 382 573 6170.1 5 3 5 6 4 10 22 13 18 25 24 21 33 43 81 83 294 310 250 270 424 383 607 6310.3 5 3 2 1 13 9 22 17 23 18 16 14 41 35 82 79 279 298 285 247 400 399 622 6600.6 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5950.8 6 7 6 2 7 7 15 7 15 32 21 9 31 34 56 53 277 321 245 246 400 387 647 5561 6 12 4 4 2 12 14 19 17 32 22 21 24 24 51 68 273 346 289 233 373 412 615 590-5 8 4 1 2 8 9 17 16 17 17 15 11 30 38 74 67 273 339 267 262 385 411 631 625-4 5 2 8 2 2 11 17 21 16 19 29 17 21 37 62 82 266 327 286 262 395 389 561 653-3 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-2 5 4 4 6 10 8 12 10 14 24 14 20 27 26 62 55 259 331 265 265 411 372 624 573-1 5 9 8 4 3 3 14 13 16 26 23 20 26 33 54 74 269 337 267 265 391 354 566 598Lane Change ParametersAccepted deceleration (own)m/s² -1TTC < 1Accepted deceleration (trailing)m/s² -1100Max deceleration (own)m/s² -4Max deceleration (Trailing)m/s² -3own (-1m/s² per distance)m 100Trailing (-1m/s² per distance)mWaiting time before diffusionSec 60Min headway (front/rear)m 0.52425Max deceleration for cooprative brakingm/s² -3Safety distance reduction factor0.6TTC < 1.5 TTC < 3#1617181920212223131 Table C.4 Sensitivity analysis of the signal control parameters and the desired deceleration Parameter Unit Default Values 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8Continuous check 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595One decision 2 3 5 3 13 6 11 14 12 18 26 16 45 36 49 85 279 313 309 247 395 426 614 644Go as green 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595Stop as red 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5950.1 21 18 17 13 30 35 49 46 37 41 32 31 46 41 66 75 187 204 181 156 215 220 299 2740.3 11 8 9 7 11 11 18 24 25 22 20 15 26 22 45 46 214 227 178 172 238 255 340 3640.6 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 5950.8 4 4 4 4 15 16 22 19 24 28 22 24 44 49 69 96 273 343 288 266 452 433 658 7511 12 0 3 2 9 9 18 14 26 27 11 25 43 31 52 72 319 340 308 291 472 479 669 7300 3 3 3 2 6 7 10 15 16 17 20 24 35 36 56 67 317 386 302 285 455 460 706 76950 1 4 3 1 14 7 19 16 16 17 27 17 33 36 69 49 283 292 275 269 401 422 630 595100 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595150 3 5 5 3 9 14 10 14 18 15 27 11 35 32 57 56 272 298 260 204 359 366 574 587200 4 8 2 2 4 10 16 11 20 22 23 6 38 33 65 72 263 300 286 219 387 380 611 6270 5 7 3 3 8 11 13 6 19 26 14 15 41 31 59 53 262 326 255 263 427 385 603 61050 3 2 6 2 6 4 10 12 24 23 23 14 33 27 44 74 273 311 260 261 372 418 574 543100 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595150 3 7 3 3 13 4 14 17 26 18 22 14 44 25 62 64 291 306 273 270 378 371 577 592200 6 5 4 4 9 5 18 13 18 13 26 11 33 41 60 57 280 296 279 277 411 424 612 587-1.2 1 6 6 0 17 5 16 16 10 17 13 8 26 25 46 73 299 317 322 247 427 439 644 693-2 5 4 4 1 3 6 14 12 17 17 14 10 22 30 54 59 283 348 313 263 409 447 621 621-2.8 10 4 2 2 4 5 12 10 19 18 21 19 38 28 64 55 254 337 286 258 416 391 599 595-3.6 10 1 8 3 10 8 15 11 25 15 22 18 36 29 71 73 228 267 246 214 383 324 564 548-4.4 2 3 3 3 7 9 26 15 25 22 18 23 39 31 64 53 247 279 252 226 374 376 548 556m/s² -2.828Reduced safety distance (Reduction factor )Decision modelBehavior at amber signal31Desired deceleration Continuous checkGo as green2627TTC < 1 TTC < 1.5 TTC < 330Reduced safety distance (end downstream)m 100#Signal Control Parameters & Desired Deceleration29Reduced safety distance (start upstream) m 1000.6132 Appendix D Computer Codes This appendix contains some of the developed computer codes used in this thesis to apply the calibration of the simulation models. D.1 MATLAB Script (Genetic Algorithm for VISSIM) % parameters ranges %-------------------------- cc0_range=0.5:0.1:2.5; cc1_range=0.5:0.1:1.3; cc4_5_range= [0.01 0.1:0.05:0.35 0.4:0.1:1.5]; SRF_range = 0.1:0.1:1; SUPD_range = 0:10:200; DD_range = -4.4:0.2:-1.2; %--*--*--*--*--*--*-- %First population %--*--*--*--*--*--*-- %Selection(use Latin Hypercube sampling to cover all space) population=[]; LHS_P = lhsdesign(10,6); for ii=1:1:10 cc0= cc0_range(max(round(LHS_P(ii,1)*length(cc0_range)),1)); cc1= cc1_range(max(round(LHS_P(ii,2)*length(cc1_range)),1)); cc4_5= cc4_5_range(max(round(LHS_P(ii,3)*length(cc4_5_range)),1)); SRF= SRF_range(max(round(LHS_P(ii,4)*length(SRF_range)),1)); SUPD= SUPD_range(max(round(LHS_P(ii,5)*length(SUPD_range)),1)); DD= DD_range(max(round(LHS_P(ii,6)*length(DD_range)),1)); pop_x = [cc0 cc1 cc4_5 SRF SUPD DD]; population=[population; pop_x]; end %%%% Write results in excel %-------------------------- headings={'CC1' 'CC3' 'CC4&CC5' 'RF' 'Fitness Value' 'min Fitness Value' 'Average Fitness Value'}; xlswrite('Calibration Results',headings,'First Population','A1'); xlswrite('Calibration Results',population,'First Population','A2'); % Simulation to calculate fitness values Total_fitness_values = []; for jjj=1:1:10 cc1=population(jjj,1); cc3=population(jjj,2); cc4=-1*population(jjj,3); cc5=population(jjj,3); RF=population(jjj,4); %%%% adjust parameters in VISSIM inp files run edit_inp_file; %%%% run VISSIM for all hours system('Calibration Number Two.exe'); 133 %%%% Calculate Fitness Value run Calculate_fitness_value xlswrite('Calibration Results',fitness_value,'First Population',['E' num2str(jjj+1)]); Total_fitness_values = [Total_fitness_values fitness_value]; Folder_new_name_pop = ['[path]... 'pop-' num2str(jjj)]; movefile([path]', ... Folder_new_name_pop); copyfile('[path]', ... '[path]’); end Min_Fitness_value = min(Total_fitness_values); Average_Fitness_value = mean(Total_fitness_values); %%%% Write results in excel %-------------------------- xlswrite('Calibration Results',Total_fitness_values','First Population','E2'); xlswrite('Calibration Results',Min_Fitness_value,'First Population','F2'); xlswrite('Calibration Results',Average_Fitness_value,'First Population','G2'); %--*--*--*--*--*--*-- % 70 Generations %--*--*--*--*--*--*-- for generation=1:1:70 %Selection New_population = []; %%% Childs by parents crossover selection (9 childs) w_fitness_values = ones(1,10)./Total_fitness_values; for child = 1:1:8 % select 2 parents parents_numbers_in_population = randsample(10,2,true,w_fitness_values); parent1= population(parents_numbers_in_population(1):parents_numbers_in_population(1),:); parent2= population(parents_numbers_in_population(2):parents_numbers_in_population(2),:); % cross over to get child min_start_ranges=min([parent1; parent2]); max_end_ranges=max([parent1; parent2]); exploration_coeff = 0.5; New_start_ranges = max([start_range;(min_start_ranges - exploration_coeff* (max_end_ranges - min_start_ranges))]); New_end_ranges = min([end_range;(max_end_ranges + exploration_coeff* (max_end_ranges - min_start_ranges))]); New_child = random('uniform',New_start_ranges,New_end_ranges); 134 New_population = [New_population; New_child]; end %%% Mutation for mutation=1 popBYmutation = random('uniform',start_range,end_range); New_population=[New_population; popBYmutation]; end %%% Elitism (keep minimum fitness value's population) Elitism_pop_number = find(Total_fitness_values<=Min_Fitness_value); popBYelitism = population(Elitism_pop_number(1), :); New_population=[New_population; popBYelitism]; %%%%.............. population = New_population; %%%% Write results in excel %-------------------------- xlswrite('Calibration Results',headings,num2str(generation),'A1'); xlswrite('Calibration Results',population,num2str(generation),'A2'); % Simulation to calculate fitness values Total_fitness_values = []; for kkk=1:1:9 cc1=population(kkk,1); cc3=population(kkk,2); cc4=-1*population(kkk,3); cc5=population(kkk,3); RF=population(kkk,4); %%%% adjust parameters in VISSIM file inp file run edit_inp_file; %%%% run VISSIM system('Calibration Number Two.exe'); %%%% Calculate Fitness Value run Calculate_fitness_value xlswrite('2Calibration Results',fitness_value,num2str(generation),['E' num2str(kkk+1)]); Total_fitness_values = [Total_fitness_values fitness_value]; Folder_new_name = ['[path]' ... num2str(generation) '-' num2str(kkk)]; movefile('[path]', ... Folder_new_name); copyfile('[path]', ... '[path]'); end Total_fitness_values = [Total_fitness_values Min_Fitness_value]; xlswrite('2Calibration Results',Min_Fitness_value,num2str(generation),'E11'); Min_Fitness_value = min(Total_fitness_values); Average_Fitness_value = mean(Total_fitness_values); %%%% Write results in excel %-------------------------- xlswrite('2Calibration Results',Min_Fitness_value,num2str(generation),'F2'); 135 xlswrite('2Calibration Results',Average_Fitness_value,num2str(generation),'G2'); end D.2 MATLAB Script (Editing of the VISSIM (*inp) File) %%%% This script to change values of cc1 cc3 cc4 cc5 SDRF in inp file %%%% Before use this script cc1 cc3 cc4 cc5 RF should be defined %%% edit RF inp_file_name = [file_path]; %%%-------------------------------------------------------------- %%% edit RF (lane change reduction factor) %%%-------------------------------------------------------------- fid1 = fopen(inp_file_name,'r+'); replaceLine = 262;%%% determine line for k=1:(replaceLine-1); fgetl(fid1);%%%% move through lines end fseek(fid1, 0, 'cof'); fprintf(fid1, '%s', [' COOPERATIVE -3.00 LOOKAHEAD ABXFACTOR ' num2str(RF,'%6.2f')]); fclose(fid1); %%%-------------------------------------------------------------- %%% edit SRF SUPD (signal reduction factor and upstream distance) %%%-------------------------------------------------------------- fid2 = fopen(inp_file_name,'r+'); replaceLine = 265;%%% determine line for k=1:(replaceLine-1); fgetl(fid2);%%%% move through lines end fseek(fid2, 0, 'cof'); fprintf(fid2, '%s',[' SIGNAL_HEAD ABXFACTOR ' num2str(SRF,'%6.2f') ' DISTANCE FROM ' num2str(SUPD,'%6.2f') ' UNTIL 100.00 ']); fclose(fid2); %%%-------------------------------------------------------------- %%% edit cc0 cc1 cc3 cc4 %%%-------------------------------------------------------------- fid3 = fopen(inp_file_name,'r+'); replaceLine = 267;%%% determine line for k=1:(replaceLine-1); fgetl(fid3);%%%% move through lines end fseek(fid3, 0, 'cof'); fprintf(fid3, '%s',[' CC0 ' num2str(cc0,'%6.2f') ' CC1 ' num2str(cc1,'%6.2f') ' CC2 4.00 CC3 ' num2str(cc3,'%6.2f') ' CC4 ' num2str(cc4,'%6.2f')]); fclose(fid3); %%%-------------------------------------------------------------- %%% edit cc5 %%%-------------------------------------------------------------- fid4 = fopen(inp_file_name,'r+'); replaceLine = 268;%%% determine line 136 for k=1:(replaceLine-1); fgetl(fid4);%%%% move through lines end fseek(fid4, 0, 'cof'); fprintf(fid4, '%s',[' CC5 ' num2str(cc5,'%6.2f') ' CC6 11.44 CC7 0.25 CC8 3.50 CC9 1.50']); fclose(fid4); %%%-------------------------------------------------------------- %%% edit DD %%%-------------------------------------------------------------- fid5 = fopen(inp_file_name,'r+'); replaceLine = 4088;%%% determine line for k=1:(replaceLine-1); fgetl(fid5);%%%% move through lines end fseek(fid5, 0, 'cof'); fprintf(fid5, '%s', [' BASE_POINT 0.000 ' num2str(DD,'%6.3f') ' ' num2str((DD-0.25),'%6.3f') ' ' num2str((DD+0.20),'%6.3f') ' 250.000 ' num2str(DD,'%6.3f') ' ' num2str((DD-0.25),'%6.3f') ' ' num2str((DD+0.20),'%6.3f')]); fclose(fid5); D.3 VISUAL BASIC Code (Automatic Runs of VISSIM Simulation) Module Module1 Sub Main() Dim vissim As VISSIM_COMSERVERLib.Vissim Dim simulation As VISSIM_COMSERVERLib.Simulation 'start VISSIM and create an instance of a Vissim object vissim = New VISSIM_COMSERVERLib.Vissim For jj = 1 To Number_of_hours 'load a network Dim fileNumber As String = Convert.ToString(jj) Dim fileName As String = \"[inp file_path]\" 'load a network vissim.LoadNet(fileName) 'random seed adjustment Dim seeds As Integer = 10 to 50 ‘3 random seeds 10,30,50 simulation = vissim.Simulation simulation.RandomSeed = seeds simulation.Speed = 0 'set evaluation files vissim.Evaluation.AttValue(\"EXPORT\") = True vissim.Evaluation.AttValue(\"TRAVELTIME\") = True vissim.Evaluation.AttValue(\"NODE\") = True 'set visualization to be hidden 137 vissim.Graphics.AttValue(\"DISPLAY\") = 2 vissim.Graphics.AttValue(\"VISUALIZATION\") = False 'save vissim file vissim.SaveNet() 'Run simulation vissim.ShowMinimized() simulation.RunContinuous() vissim.SaveNet() Next End Sub End Module D.4 MATLAB Script (Genetic Algorithm for PARAMICS) %% Ranges of parameters MTH_range = [0.4 0.8 1 1.4 2.4]; MRT_range = [0.3 0.5 1 1.6 2.4]; MinG_range = [1.5 2 2.5 3]; %--*--*--*--*--*--*-- %First population %--*--*--*--*--*--*-- Number_of_first_pop =10; %Selection(use Latin Hypercube sampling to cover all space) population=[]; LHS_P = lhsdesign(Number_of_first_pop,3); samples_MTH = round(LHS_P(:,1)*5); samples_MRT = round(LHS_P(:,2)*5); samples_MinG = round(LHS_P(:,3)*4); for pop=1:1:Number_of_first_pop if samples_MTH(pop) ==0 samples_MTH(pop) = 1; end if samples_MRT(pop) == 0 samples_MRT(pop) = 1; end if samples_MinG(pop) == 0 samples_MinG(pop) = 1; end MTH = MTH_range(samples_MTH(pop)); MRT = MRT_range(samples_MRT(pop)); MinG = MinG_range(samples_MinG(pop)); pop_x = [MTH MRT MinG]; %%%% to get values of parameters corresponding to LHS CDF population=[population; pop_x]; end %%%% Write results in excel 138 %-------------------------- headings={'MHT' 'MRT' 'MinG' 'Fitness Value' 'min Fitness Value' 'Average Fitness Value'}; xlswrite('GA Results',headings,'First Population','A1'); xlswrite('GA Results',population,'First Population','A2'); % Simulation to calculate fitness values Total_fitness_values = []; for jjj=1:1:Number_of_first_pop MTH=population(jjj,1); MRT=population(jjj,2); MinG=population(jjj,3); %%%% Calculate Fitness Value run Calculate_fitness_value Total_fitness_values = [Total_fitness_values fitness_value]; end xlswrite('GA Results',Total_fitness_values','First Population','D2'); Min_Fitness_value = min(Total_fitness_values) Average_Fitness_value = mean(Total_fitness_values); Total_min_fitness_values = Min_Fitness_value; Total_average_fitness_values = Average_Fitness_value; %--*--*--*--*--*--*-- % 40 Generations %--*--*--*--*--*--*-- for generation=1:1:40 %Selection New_population = []; %%% Childs by parents crossover selection (8 childs) w_fitness_values = ones(1,Number_of_first_pop)./Total_fitness_values; for child = 1:1:(Number_of_first_pop-2) % select 2 parents parents_numbers_in_population = randsample(Number_of_first_pop,2,true,w_fitness_values); %% Wieghted random selection parent1= population(parents_numbers_in_population(1):parents_numbers_in_population(1),:); parent2= population(parents_numbers_in_population(2):parents_numbers_in_population(2),:); % cross over to get child (using ranges instead of one value) Two_parents=[parent1; parent2]; New_child = [Two_parents(randsample(2,1),1) ... Two_parents(randsample(2,1),2) ... Two_parents(randsample(2,1),3)]; New_population = [New_population; New_child]; end %%% Mutation (1 child) for mutation=1 popBYmutation = [MTH_range(randsample(5,1))... MRT_range(randsample(5,1)) ... MinG_range(randsample(4,1))]; 139 New_population=[New_population; popBYmutation]; end %%% Elitism (keep minimum fitness value's population)(1 child) Elitism_pop_number = find(Total_fitness_values<=Min_Fitness_value); popBYelitism = population(Elitism_pop_number(1), :); New_population=[New_population; popBYelitism]; %%%%.............. population = New_population; %%%% Write results in excel %-------------------------- xlswrite('GA Results',headings,num2str(generation),'A1'); xlswrite('GA Results',population,num2str(generation),'A2'); % Simulation to calculate fitness values Total_fitness_values = []; for kkk=1:1:Number_of_first_pop MTH=population(kkk,1); MRT=population(kkk,2); MinG=population(kkk,3); %%%% Calculate Fitness Value run Calculate_fitness_value Total_fitness_values = [Total_fitness_values fitness_value]; end xlswrite('GA Results',Total_fitness_values',num2str(generation),'D2'); Min_Fitness_value = min(Total_fitness_values) Average_Fitness_value = mean(Total_fitness_values); Total_min_fitness_values = [Total_min_fitness_values Min_Fitness_value]; Total_average_fitness_values = [Total_average_fitness_values Average_Fitness_value]; end %%%% Write results in excel %-------------------------- xlswrite('GA Results',{'Generation' 'Min Fitness Value' 'Average Fitness Value'},'Final Results','A1'); xlswrite('GA Results',[1:1:(generation+1)]','Final Results','A2'); xlswrite('GA Results',Total_min_fitness_values','Final Results','B2'); xlswrite('GA Results',Total_average_fitness_values','Final Results','C2'); 140 D.5 MATLAB Script (Editing of the PARAMICS Configuration File and Run Simulation) importfile1('[Configuration file full path]'); Speed_memory = round(1.5*MRT*10); line9_text =['speed memory ' num2str(Speed_memory)]; configuration{9,1} = line9_text; line21_text =['mean headway ' num2str(MTH)]; configuration{21,1} = line21_text; line22_text =['mean reaction time ' num2str(MRT)]; configuration{22,1} = line22_text; line23_text =['gap ' num2str(MinG) ' m']; configuration{23,1} = line23_text; x=char(configuration); dlmwrite('[New Configuration file full path]', x, ''); dos 'processor-cmd -cmd -netpath AfterFirstCalibration -file configuration' D.6 MATLAB Function (Check Traffic Signal Indications) function [ color ] = check_color(x) %CHECK_COLOR Summary of this function goes here % x is a picture (frame want to be checked) y_red = x(170:172,74:76,1:3); y_yellow = x(188:190,76:78,1:3); y_green = x(203:205,79:81,1:3); y_green_left = x(220:221,81:83,1:3); if min(min(y_red(:,:,1)))>100 if min(min(y_green_left(:,:,2)))>100 && max(max(y_green_left(:,:,2)))>200 color='red & green left'; else color='red'; end elseif min(min(y_yellow(:,:,2)))>100 color='yellow'; elseif min(min(y_green(:,:,2)))>100 if min(min(y_green_left(:,:,2)))>100 && max(max(y_green_left(:,:,2)))>200 color='green & green left'; else color='green'; end else color='none'; end end "@en ; edm:hasType "Thesis/Dissertation"@en ; vivo:dateIssued "2015-11"@en ; edm:isShownAt "10.14288/1.0165811"@en ; dcterms:language "eng"@en ; ns0:degreeDiscipline "Civil Engineering"@en ; edm:provider "Vancouver : University of British Columbia Library"@en ; dcterms:publisher "University of British Columbia"@en ; dcterms:rights "Attribution-NonCommercial-NoDerivs 2.5 Canada"@* ; ns0:rightsURI "http://creativecommons.org/licenses/by-nc-nd/2.5/ca/"@* ; ns0:scholarLevel "Graduate"@en ; dcterms:title "Calibration and validation of traffic microsimulation models for safety evaluation using automated video-based conflict analysis"@en ; dcterms:type "Text"@en ; ns0:identifierURI "http://hdl.handle.net/2429/54854"@en .