International Construction Specialty Conference of the Canadian Society for Civil Engineering (ICSC) (5th : 2015)

Research for generating 3D model from laser scanner data removed noise Tanaka, Shigenori; Imai, Ryuichi; Nakamura, Kenji; Kawano, Kouhei; Kubota, Satoshi Jun 30, 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
52660-Tanaka_S_et_al_ICSC15_319_Research_For_Generating.pdf [ 1.85MB ]
52660-Tanaka_S_et_al_ICSC15_319_Research_For_Generating_slides.pdf [ 2.1MB ]
Metadata
JSON: 52660-1.0076443.json
JSON-LD: 52660-1.0076443-ld.json
RDF/XML (Pretty): 52660-1.0076443-rdf.xml
RDF/JSON: 52660-1.0076443-rdf.json
Turtle: 52660-1.0076443-turtle.txt
N-Triples: 52660-1.0076443-rdf-ntriples.txt
Original Record: 52660-1.0076443-source.json
Full Text
52660-1.0076443-fulltext.txt
Citation
52660-1.0076443.ris

Full Text

5th International/11th Construction Specialty Conference 5e International/11e Conférence spécialisée sur la construction    Vancouver, British Columbia June 8 to June 10, 2015 / 8 juin au 10 juin 2015   RESEARCH FOR GENERATING 3D MODEL FROM LASER SCANNER DATA REMOVED NOISE Shigenori Tanaka1,7, Ryuichi Imai2,3, Kenji Nakamura4 , Kouhei Kawano5  and Satoshi Kubota6 1 Faculty of Informatics, Kansai University, Japan 2 National Institute for Land and Infrastructure Management, MLIT, Japan 3 Cooperative Graduate School of Informatics, Kansai University, Japan 4 Faculty of Information Technology and Social Sciences, Osaka University of Economics, Japan 5 Graduate School of Informatics, Kansai University, Japan 6 Environmental and Urban Engineering, Kansai University, Japan 7 tanaka@res.kutc.kansai-u.ac.jp Abstract: Construction CALS/EC is introduced in public works projects over the whole life cycle for the purpose of reduction of the construction period, quality guarantee, and cost reduction.  Especially recently, environmental improvement of the information-oriented construction using 3D data attracts attention for engineering works stage.  In the construction site, it is expected to create 3D model from the point cloud data obtained by surveying the site with the total station or the laser scanner.  However, it is necessary to solve the problem of how to handle the great amount of point cloud data obtained at the survey based on their characteristics, as well as the problem of required accuracy for the information-oriented construction.  The authors have devised a technique to make 3D model of the river embankment, which satisfy the required accuracy of the information-oriented construction from a large quantity of point cloud data.  This technique used the point cloud data of the river embankment measured with the laser scanner and DM data of it.  Then, the boundary line (breakline) between the crown surface and the slope face of the river embankment is extracted automatically to create 3D model on CAD.  However, the following problems were revealed; the freshness depending on the update cycle of DM data, and wrong extraction of a breakline due to the noise such as the vegetation between the crown surface and the slope face of the river embankment.  In this research, we devised a method for automatically extracting a breakline by inferring the crown surface area of the river embankment from the point cloud data, and a method for removing any point cloud data that are the noise such as the vegetation on the crown surface of the embankment.  Then we performed evaluation experiments and proved the usability of the devised solution. 1 INTRODUCTION Public institutions in Japan are promoting measures for constructing a data circulation environment using ICT  (Information and Communication Technology) over the whole life cycle for reduction of construction period, quality assurance, and cost reduction of public works.  Recently, there are efforts underway by academy, industry, and government to apply this in product models in construction stages such as development of Road Alignment Data Exchange Standard, as well as advanced efforts for intelligent construction (Tanaka, 2009). 319-1 Against this background, a workshop on establishing a data circulation environment for river projects was set up in Kinki Regional Development Bureau, Ministry of Land, Infrastructure, Transport and Tourism for the purpose of promoting new technology development with public and academic sectors as its core(Fukumori, 2009).  This workshop aims to develop technologies for generating high-precision three-dimensional (3D) models useful in construction management or maintenance using the survey data concerning the present topography of rivers.  Relevant researches suggest methods of generating 3D models automatically from the point cloud data(Shaohui, 2013)(Choi, 2008)(Tanaka, 2010) measured with high-precision scanners(Ohtu, 2007).  However, there are two potential problems in the 3D models created from the point cloud data collected by the laser scanner: how to handle a large amount of point cloud data, and wrong extraction of the boundary line of aspect change (hereinafter referred to as 'breakline'). To solve these problems, the existing researches propose a method to extract breaklines, leaving geometric characteristics and thinning out an enormous amount of point cloud data, through a method of using relative positions of point cloud data(Shaohui, 2013), or a method of using reflection intensity of point cloud data (Choi, 2008).  However, it is impossible to uniquely identify the boundary lines between planes of the river embankment because it is a space where artifacts and natural objects coexist.  Accordingly, it is difficult to create a 3D model with high precision of reproducibility.  Therefore we have been working on the research on controlling wrong extraction of breaklines(Tanaka, 2010).  Firstly in specific, point cloud data was superimposed on DM data developed nationwide by public survey.  Next, a 3D model was automatically generated using the road alignment contained in DM data as a clue.  However, there are two problems of "Freshness due to the update cycle of DM data" and "That noise such as vegetation inhabiting the top plane of a river levee (hereinafter referred to as "crown surface")". Based on these problems, this research aims to establish a method of automatically generating a 3D model from a large amount of point cloud data in consideration of boundaries of planes of the river levee.  This method assumes the outline of the crown surface estimated from the point cloud data as a candidate breakline.  Furthermore, wall-shaped noise is eliminated from the point cloud data on the crown surface automatically to restraint wrong extraction of breaklines. 2 OVERVIEW OF OUR RESEARCH This research proposes an automatic generation method of 3D models that solves two problems: "problem of freshness due to the update cycle of DM data" and "problem of wrong extraction of breaklines because the boundary between planes becomes vague due to the wall-shaped noise". 2.1 Measures against the problem of freshness due to the update cycle of DM data This method identifies an area showing the crown surface from point cloud data.  The outline of the identified area is assumed the candidate breakline.  Therefore, a 3D model is constructed from point cloud data without depending on DM data. 2.2 Measures against the problem of wrong extraction of breaklines because the boundary between planes becomes vague due to the wall-shaped noise This method estimates ambiguous ground surface from features of the shape and the noise of the river embankment. 2.3 Processing flow Figure 1 shows the processing flow proposed in this research.  This research adds a function of generating candidate breaklines and a removal processing of wall-shaped noise to the existing method(Tanaka, 2010).  Thus the proposed method is comprised of three functions; candidate breakline generation, breakline extraction, and thinning out point cloud data.  The processing procedure of each function is shown below. 319-2 Function of candidate breaklinegenerationProcess of dividing elevationProcess of clustering densityProcess of identification crown surfaceProcess of limiting the extraction rangeFunction of breaklines extractionProcess of generating cross-section modelProcess of removal wall-shaped noiseProcess of identification acquiringcross-sectional change pointsProcess of creating breaklinesInputFunction of thinning out point cloud dataProcess of creating a filterProcess of interpolation of the point cloud dataPoint cloud dataInputCandidate breaklineOutputBreaklinedataOutput InputThinned-out point cloud dataOutput Figure 1: Processing flow The function of candidate breakline generation is comprised of three kinds of processing: process of dividing elevation, process of clustering density, and process of identification crown surface.  Process of dividing elevation is the process of entering the point cloud data and divided them into multiple layers using elevation values.  In order to distinguish the layer on which the point cloud data are concentrated, the point cloud data on each layer is clustered based on the relative distance between points, and a cluster of micro area is removed as noise.  Process of identification crown surface is the process of evaluating the density of point cloud data of each layer to estimate the crown surface layer.  Then the outline of the point cloud data region contained within the crown surface is obtained as a candidate breakline.  Function of breaklines extraction comprises 5 kinds of processing; limiting the extraction range, generating a cross-section model, removing wall-shaped noise, identifying the cross-sectional change point, and creating breaklines.  Process of limiting the extraction range limits the extraction range of the cross-sectional change points using the candidate breaklines acquired from the point cloud data and the candidate breakline acquisition function.  Process of generating cross-section model acquires cross section models of the levee from the point cloud data contained in the extraction range.  Wall-shaped noise removal processing removes the point cloud data with the altitude higher than the crown surface and acquires cross section models from which the influence of noise has been removed.  Identification processing of cross-sectional change points acquires an intersection point of the line segments showing the crown surface and the slope from the cross section model from which wall-shaped noise has been removed as the cross-sectional change point.   Process of creating breaklines creates a breakline by connecting cross-sectional change points through process of identification acquiring cross-sectional change point.  Function of thinning out point cloud data comprises process of creating a filter and process of interpolation of the point cloud data.  Process of creating a filter creates a breakline filter by superimposing breaklines and a grid-like filter.  Next, process of interpolation of point cloud data interpolates the point cloud data using the breakline filter.  This process converts the point cloud data measured at irregular intervals to the thinned-out point cloud data in a grid-like shape with breaklines taken into consideration. 3 FUNCTION OF CANDIDATE BREAKLINE GENERATION The function of candidate breakline generation creates candidate breaklines from the point cloud data.  Here we assume the point cloud data measured by the laser scanner as P={p1, p2, p3, ..., pi}.  This function solves the aforementioned problem of freshness due to the update cycle of DM data. 3.1 Process of dividing elevation This processing divides point cloud data into multiple layers as pre-processing to distinguish the region of the crown surface from the point cloud data.  Specifically, the point cloud data P is divided at regular intervals at every height h, as shown in Figure 2.  The layer set acquired from this processing is assumed as L={l1, l2, l3, ..., lj}. 319-3 XZYXZYhXZYl1l2l3Point cloud data P Divided at regular intervals h Divided layer L  Figure 2: Process of dividing elevation 3.2 Process of clustering density In order to distinguish the layer on which point cloud data are concentrated, this processing performs clustering by the density of points over the layer and calculates the region area of each cluster.  The characteristics of the point cloud data obtained by measuring a river levee with MMS shows that the density of the point cloud data of the crown surface close to the measurement vehicle tends to be high, and that the point cloud data density of those away from the measurement vehicle such as slope or foot of slope tends to be low.  For this reason, this processing used DBSCAN method, which is one of the data mining methods, as a method for clustering based on density.  DBSCAN is a clustering method based on the point density using two values: distance threshold Eps, and threshold of target numbers MinPts.  Figure 3 shows a method of process of clustering density.  First, the point cloud data contained in an arbitrary layer l j, Pdbscan={pd(j, 1), pd(j, 2), pd(j, 3), ..., pd(j, l)}, is clustered by DBSCAN.  When arbitrary point pd(j, l) in the point cloud data Pdbscan and neighborhood point pd(j, l+1) contained in the neighborhood point cloud NEps(pd(j, l)) within the distance Eps from that point satisfy the following conditional equations [1] and [2], the neighborhood point pd(j, l+1) is classified as the same cluster with the point pd(j, l). [1]  )( )1,(),( +∈ ljEpslj pdNpd   [2]  MinPtspdN ljEps ≥+ )( )1,( Assume the clustering result by DBSCAN from an arbitrary layer l j as Clj={cl(j, 1), cl(j, 2), cl(j, 3), ..., cl(j, k)}.  Then obtain the region outline of each cluster cl(j, k) of the cluster set Clj by the active contour method.  In addition, assume the point cloud data applied to the active contour method as Psnake={ps(j, 1), ps(j, 2), ps(j, 3)}.  Region area Sl of the cluster set ClPoint cloud data of Layer ljXYCluster set Cl of layer ljXYXY Cluster cl(j, 1)Area sl(j, 1)Cluster cl(j, 2)Area sl(j, 2)Cluster cl(j, 3)Area sl(j, 3) Figure 3: Process of clustering density Figure 4 shows the outline of the active contour method used in this research.  The active contour method is a method for capturing the contour of an object by repeating movement and deformation of the closed curves that contain the object.  First, create a circle so that it circumscribes the target point cloud data Psnake.  Second, create point-series Ptangent={pt1, pt2, pt3, ..., ptn} arranged at equal intervals on the 319-4 circumference, and a tangent line of the circle T={t1, t2, t3, ..., tn} passing through ptn.  Then, translate the tangent line of the circle tn passing through the point ptn on the circumference towards the center point of the circle until it overlaps with the target point cloud data ps(j, m).  Then the distance translated by ptn,  range is calculated using equation [3].  Here tn is a straight line formula ax+by+c=0.  xt and yt are assumed as x and y coordinates of the translating point ptn, whereas xs and ys as x and y coordinates of the target point cloud data ps(j, m). [3]  22 bacybxacybxaRange sstt++⋅+⋅−+⋅+⋅=  Point-series placed at regular intervals on the circumference PtangentPoint cloud dataXYXYObtaining a set of tangent lines T’ of point cloud dataT’XYTranslating the tangent line tn of the circleXY tn Figure 4: Outline of the active contour method used in this study The set of tangent lines of point cloud data obtained by this processing is assumed as T'={t'1, t'2, t'3, ..., t'n}.  The set of intersections of a tangent line set T' is assumed as the region outline F={f1, f2, f3, ..., fo}.  Then the region area Slj={sl(l, 1), sl(l, 2), sl(l, 3), ..., sl(l, k)} of the cluster set Clj  is calculated from the region outline F of the cluster obtained by the active contour method.  Region area Slj of each cluster is calculated by calculating the area of the following polygons using equation [4]. xo and yo in equation [4] are x and y coordinates of vertex fo of the region outline F.  xo+1 and yo+1 are x and y coordinates of the point fo+1 adjacent to vertex fo. [4] ∑=++ +⋅−=Foooookj yyxxsl111),( )()(21  3.3 Process of identification crown surface Figure 5 shows the process of identification crown surface.  The process of identification crown surface identifies the layer of the crown surface in order to acquire the candidate breakline.  The layer of the crown is estimated using the ratio of the region area of the point cloud data belonging to the layer to the sum total of region area of each cluster Slj, or the region size where there are point cloud data.   Then, the region area of each layer is assumed as S={s1, s2, s3, ..., sj}.  These region areas are different from layer to layer since they depend on the result of clustering processing of each layer. First, when it is a micro cluster of which the region area is threshold MinSize or smaller, the micro cluster is removed from any layer l j.  Next, the region outline and region area of any layer l j is calculated using the active contour method like the preceding term.  Then using the region area of the layer and the region area of the cluster contained in the layer, layer lj that is to be the crown surface is identified using equation [5].  Finally, the region outline of the layer l j of the crown surface is obtained as the candidate breakline BL. 319-5 Table 1: Experiment environment Type Specifications Hardware Experiment equipment CPU Intel® Core™2 Duo CPU 2.50Ghz Memory 4.0GB HDD 280GB Software Developing  environment and language Visual Studio 2010 Visual C# CAD AutoCAD Civil3D 2010  Table 2: Details of the experimental data Items MMS data Number of points About 4.1million Measuring distance 800m Absolute precision Horizontal direction 10cm Vertical direction 15cm Relative precision 1cm  4.3 Experiment description First, three 3D models are generated by entering the point cloud data. • Existing method (Method a) (Tanaka, 2010) • Proposed method with a function of candidate breakline generation added to the existing method (Method b) • Proposed method with a function of candidate breakline generation and process of removal wall-shaped noise added to the existing method (Method c) Next, cross sections are obtained from the generated 3D model.  Using the longitude and latitude of the distance marks in the measured cross sections of correct data, information on the same place is extracted as to cross sections.  Finally, the cross sections is compared with their correct data for evaluating the precision. 4.4 Experimental results and discussion This experiment evaluates the precision using the difference in elevation of each cross section by superposing the cross section of a 3D model created through each method with the measured cross section.  For the measured cross sections to be used in this experiment, 5 drawings were prepared at every 200m between the distance marks of 7.2km and 8.0km points contained within the measurement area using MMS.  Figure 8 shows an image of comparing cross sections.  Here comparison of the difference in elevation is calculated over the evaluation points created by dividing the measured cross section at intervals 1cm.  Moreover, the comparison results are classified into three sections: all, section A and section B, and totalized as shown in Figure 9. Difference in elevationElevation (m)Distance (m)Cross section of the 3D modelMeasured cross section Figure 8: Image of comparing cross sections Overall Section A5cm 5cmSection B Section B  Figure 9: Groups of totalizing comparison results Section A is assumed as the range including the width of the crown surface with 5cm added on both sides.  This is determined with reference to the value of the work progress control standard for filling works of river earthwork providing the accidental error for the crown surface should be within 10cm. Section B is assumed as the range not contained in Section A except the crown surface.  Table 3 shows the totalized results for 3D models created using each method and the measured cross section drawing.  319-7 This experiment totalized to what degree the difference in elevation between cross sections is contained within 15cm at intervals of 5cm.  This is a reason there is absolute error of 15cm (in the vertical direction) in the measurement data using MMS.  Furthermore, for each totalized section, underlines were added to the values with the lowest error between the cross sections created using three methods and the measured cross section drawings.  The experimental results proved the following three points. Table 3: Comparison results between 3D models and measured cross sections Measured points Overall (%) Section A (%) Section B (%) Method  a Method  b Method  c Method  a Method  b Method  c Method  a Method  b Method  c Distance mark  7.2km ~ 5cm 22.68  20.45  23.64  64.86  81.08  91.89  17.03  12.32  14.49  ~10cm 29.71  29.07  37.38  72.97  100.00  100.00  23.91  19.57  28.99  ~15cm 35.46  33.23  45.37  78.38  100.00  100.00  29.71  24.28  38.04  Distance mark  7.4km ~ 5cm 6.33  10.67  8.33  0.00  0.00  0.00  6.99  11.76  9.19  ~10cm 16.00  25.33  21.67  0.00  0.00  3.57  17.65  27.94  23.53  ~15cm 29.00  41.33  49.33  60.71  7.14  75.00  31.99  44.85  46.69  Distance mark  7.6km ~ 5cm 2.33  2.33  2.84  0.00  0.00  2.44  2.69  2.69  2.99  ~10cm 5.43  5.43  15.50  0.00  0.00  90.24  6.29  6.29  6.89  ~15cm 11.37  8.79  20.67  12.20  0.00  100.00  11.68  10.18  11.68  Distance mark  7.8km ~ 5cm 23.49  16.95  24.94  53.06  0.00  73.47  19.51  19.23  18.41  ~10cm 37.29  37.77  37.77  81.63  81.63  91.84  31.32  31.87  30.49  ~15cm 50.36  50.85  49.15  100.00  100.00  100.00  43.68  44.23  42.31  Distance mark  8.0km ~ 5cm 15.38  15.38  21.47  48.94  59.57  85.11  9.43  7.55  10.19  ~10cm 27.88  27.56  33.65  89.36  100.00  100.00  16.98  14.72  21.89  ~15cm 53.21  50.96  54.49  100.00  100.00  100.00  44.91  42.26  46.42  Overall average ~ 5cm 14.04  13.15  16.25  33.37  28.13  50.58  11.13  10.71  11.05  ~10cm 23.26  25.03  29.20  48.79  56.33  77.13  19.23  20.08  22.36  ~15cm 35.88  37.03  43.80  70.26  61.43  95.00  32.39  33.16  37.03  First, the total average in Table 3 reveals that the proposed method c realized the smallest errors compared with the measured cross sections throughout the overall part, Section A, and Section B.  The average of the total results for Section A reveals that about 95% is included within 15cm, which is the allowable error for MMS, indicating that a 3D model with high precision has been generated.  This is a reason the vegetation noise in the upper part of the crown surface is properly removed. Second, Section A with the distance mark of 7.6km in Table 3 reveals that the 3D model generated by Method c is able to reproduce the crown surface more precisely than the ones created by the other two methods.  Figure 10 shows the result of superposing the cross section at the distance mark of 7.6km with cross section drawings created from the 3D models by each method.  Figure 11 reveals that the cross sections of 3D models created by Methods a and b do not agree with the measured cross sections in the part of crown surface, indicating that the 3D models are generated at higher position than the measured cross section.  On the other hand, the 3D model created by Method c is able to have represented the present shape, showing the similar shape of the crown surface as that of the measured cross section.  When analyzing the point cloud data at the distance mark 7.6km, there were wall-shaped noises at the upper part of crown surface including the top of slope.  On the other hand, the 3D model generated using the proposed method was able to remove the wall-shaped noise on the crown surface with high precision. 319-8 In this research, all the point cloud existing on the crown surface was assessed as noise in order to estimate the boundary lines of vague planes represented by point cloud data.  In the future, demonstration experiments will be conducted to verify to what degree of differences occur between the 3D model generated using such a noise removal method and actual present topographic features, and the applicability to practices and effects to be enjoyed will be made clear. This research generated 3D models for a river levee, and since the proposed method of this researchhas high versatility, its applicability to other civil engineering structures (roads, tunnels, or dams) will be examined in the future.  Moreover, though the applicability was verified with focus on construction stages (information-oriented construction), some of the challenges to solve in future can be development into a method for assisting creation of register maps for maintenance as well as a management approach using spatial-temporal model added with time base. Finally, this paper is part of the outcome of the activities of "Workshop on establishing a data circulation environment for river projects" set up in Kinki Regional Development Bureau, MLIT for the purpose of promoting new technology development with public and academic sectors as its core.  We plan to make further studies to realize environment for smooth data circulation. Acknowledgements We would like to thank the members at MLIT Kinki Regional Development Bureau MILT for their Invaluable advice given on this paper, Nippon Koei Co.,Ltd.  and Mitsubishi Electric Corporation for their cooperation in measurement of the point cloud data used in this research. References Tanaka, S., Imai, R. and Nakamura, K. : Examination of Framework for the Development of Data Distribution Environment for Public Works, Journal of Applied Computing in Civil Engineering, Japan Society of Civil Engineers, Vol.18, pp.37-46, 2009. (In Japanese) Fukumori, H., Sada, T. and Okubo, H. : Study on the Measurement Accuracy of 3D Laser scanner, Journal of Applied Computing in Civil Engineering, Japan Society of Civil Engineers, Vol.18, pp.193-200, 2009. (In Japanese) Ohtu, S. and Sada, T. : Method of Processing Data with a Cloud of Points on the 3-D Shape Surveying, Journal of Applied Computing in Civil Engineering, Japan Society of Civil Engineers, Vol.16, pp.27-36, 2007. (In Japanese) Taniguchi, T. and Mashita, K. : Generation of 3-Dimensional Domain from Point Clouds and Its Finite Element Modeling, Bulletin of the Japan Society for Industrial and Applied Mathematics, The Japan Society for Industrial and Applied Mathematics, Vol.15, pp.310-319, 2005. (In Japanese) Shaohui, S. and Carl, S. : Aerial 3D Building Detection and Modeling From Airborne LiDAR Point Clouds, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, IEEE, Vol. 6, No.3, pp.1440-1449, 2013. Choi, Y., Jang, Y., Lee, H. and Cho, G. : Three-Dimensional LiDAR Data Classifying to Extract Road Point in Urban Area, IEEE Geosciences and Remote Sensing Letters, IEEE, Vol.5, No.4, pp.725-729, 2008. Tanaka, S., Imai, R., Nakamura, K. and Kawano, K. :Research on Generat       Breakline from Point Cloud Data, Proceedings of the 10th International Conference on Construction Applications of Virtual Reality, CONVR2010, pp.347-356, 2010. 319-10  5th International/11th Construction Specialty Conference 5e International/11e Conférence spécialisée sur la construction    Vancouver, British Columbia June 8 to June 10, 2015 / 8 juin au 10 juin 2015   RESEARCH FOR GENERATING 3D MODEL FROM LASER SCANNER DATA REMOVED NOISE Shigenori Tanaka1,7, Ryuichi Imai2,3, Kenji Nakamura4 , Kouhei Kawano5  and Satoshi Kubota6 1 Faculty of Informatics, Kansai University, Japan 2 National Institute for Land and Infrastructure Management, MLIT, Japan 3 Cooperative Graduate School of Informatics, Kansai University, Japan 4 Faculty of Information Technology and Social Sciences, Osaka University of Economics, Japan 5 Graduate School of Informatics, Kansai University, Japan 6 Environmental and Urban Engineering, Kansai University, Japan 7 tanaka@res.kutc.kansai-u.ac.jp Abstract: Construction CALS/EC is introduced in public works projects over the whole life cycle for the purpose of reduction of the construction period, quality guarantee, and cost reduction.  Especially recently, environmental improvement of the information-oriented construction using 3D data attracts attention for engineering works stage.  In the construction site, it is expected to create 3D model from the point cloud data obtained by surveying the site with the total station or the laser scanner.  However, it is necessary to solve the problem of how to handle the great amount of point cloud data obtained at the survey based on their characteristics, as well as the problem of required accuracy for the information-oriented construction.  The authors have devised a technique to make 3D model of the river embankment, which satisfy the required accuracy of the information-oriented construction from a large quantity of point cloud data.  This technique used the point cloud data of the river embankment measured with the laser scanner and DM data of it.  Then, the boundary line (breakline) between the crown surface and the slope face of the river embankment is extracted automatically to create 3D model on CAD.  However, the following problems were revealed; the freshness depending on the update cycle of DM data, and wrong extraction of a breakline due to the noise such as the vegetation between the crown surface and the slope face of the river embankment.  In this research, we devised a method for automatically extracting a breakline by inferring the crown surface area of the river embankment from the point cloud data, and a method for removing any point cloud data that are the noise such as the vegetation on the crown surface of the embankment.  Then we performed evaluation experiments and proved the usability of the devised solution. 1 INTRODUCTION Public institutions in Japan are promoting measures for constructing a data circulation environment using ICT  (Information and Communication Technology) over the whole life cycle for reduction of construction period, quality assurance, and cost reduction of public works.  Recently, there are efforts underway by academy, industry, and government to apply this in product models in construction stages such as development of Road Alignment Data Exchange Standard, as well as advanced efforts for intelligent construction (Tanaka, 2009). 319-1 Against this background, a workshop on establishing a data circulation environment for river projects was set up in Kinki Regional Development Bureau, Ministry of Land, Infrastructure, Transport and Tourism for the purpose of promoting new technology development with public and academic sectors as its core(Fukumori, 2009).  This workshop aims to develop technologies for generating high-precision three-dimensional (3D) models useful in construction management or maintenance using the survey data concerning the present topography of rivers.  Relevant researches suggest methods of generating 3D models automatically from the point cloud data(Shaohui, 2013)(Choi, 2008)(Tanaka, 2010) measured with high-precision scanners(Ohtu, 2007).  However, there are two potential problems in the 3D models created from the point cloud data collected by the laser scanner: how to handle a large amount of point cloud data, and wrong extraction of the boundary line of aspect change (hereinafter referred to as 'breakline'). To solve these problems, the existing researches propose a method to extract breaklines, leaving geometric characteristics and thinning out an enormous amount of point cloud data, through a method of using relative positions of point cloud data(Shaohui, 2013), or a method of using reflection intensity of point cloud data (Choi, 2008).  However, it is impossible to uniquely identify the boundary lines between planes of the river embankment because it is a space where artifacts and natural objects coexist.  Accordingly, it is difficult to create a 3D model with high precision of reproducibility.  Therefore we have been working on the research on controlling wrong extraction of breaklines(Tanaka, 2010).  Firstly in specific, point cloud data was superimposed on DM data developed nationwide by public survey.  Next, a 3D model was automatically generated using the road alignment contained in DM data as a clue.  However, there are two problems of "Freshness due to the update cycle of DM data" and "That noise such as vegetation inhabiting the top plane of a river levee (hereinafter referred to as "crown surface")". Based on these problems, this research aims to establish a method of automatically generating a 3D model from a large amount of point cloud data in consideration of boundaries of planes of the river levee.  This method assumes the outline of the crown surface estimated from the point cloud data as a candidate breakline.  Furthermore, wall-shaped noise is eliminated from the point cloud data on the crown surface automatically to restraint wrong extraction of breaklines. 2 OVERVIEW OF OUR RESEARCH This research proposes an automatic generation method of 3D models that solves two problems: "problem of freshness due to the update cycle of DM data" and "problem of wrong extraction of breaklines because the boundary between planes becomes vague due to the wall-shaped noise". 2.1 Measures against the problem of freshness due to the update cycle of DM data This method identifies an area showing the crown surface from point cloud data.  The outline of the identified area is assumed the candidate breakline.  Therefore, a 3D model is constructed from point cloud data without depending on DM data. 2.2 Measures against the problem of wrong extraction of breaklines because the boundary between planes becomes vague due to the wall-shaped noise This method estimates ambiguous ground surface from features of the shape and the noise of the river embankment. 2.3 Processing flow Figure 1 shows the processing flow proposed in this research.  This research adds a function of generating candidate breaklines and a removal processing of wall-shaped noise to the existing method(Tanaka, 2010).  Thus the proposed method is comprised of three functions; candidate breakline generation, breakline extraction, and thinning out point cloud data.  The processing procedure of each function is shown below. 319-2 Function of candidate breaklinegenerationProcess of dividing elevationProcess of clustering densityProcess of identification crown surfaceProcess of limiting the extraction rangeFunction of breaklines extractionProcess of generating cross-section modelProcess of removal wall-shaped noiseProcess of identification acquiringcross-sectional change pointsProcess of creating breaklinesInputFunction of thinning out point cloud dataProcess of creating a filterProcess of interpolation of the point cloud dataPoint cloud dataInputCandidate breaklineOutputBreaklinedataOutput InputThinned-out point cloud dataOutput Figure 1: Processing flow The function of candidate breakline generation is comprised of three kinds of processing: process of dividing elevation, process of clustering density, and process of identification crown surface.  Process of dividing elevation is the process of entering the point cloud data and divided them into multiple layers using elevation values.  In order to distinguish the layer on which the point cloud data are concentrated, the point cloud data on each layer is clustered based on the relative distance between points, and a cluster of micro area is removed as noise.  Process of identification crown surface is the process of evaluating the density of point cloud data of each layer to estimate the crown surface layer.  Then the outline of the point cloud data region contained within the crown surface is obtained as a candidate breakline.  Function of breaklines extraction comprises 5 kinds of processing; limiting the extraction range, generating a cross-section model, removing wall-shaped noise, identifying the cross-sectional change point, and creating breaklines.  Process of limiting the extraction range limits the extraction range of the cross-sectional change points using the candidate breaklines acquired from the point cloud data and the candidate breakline acquisition function.  Process of generating cross-section model acquires cross section models of the levee from the point cloud data contained in the extraction range.  Wall-shaped noise removal processing removes the point cloud data with the altitude higher than the crown surface and acquires cross section models from which the influence of noise has been removed.  Identification processing of cross-sectional change points acquires an intersection point of the line segments showing the crown surface and the slope from the cross section model from which wall-shaped noise has been removed as the cross-sectional change point.   Process of creating breaklines creates a breakline by connecting cross-sectional change points through process of identification acquiring cross-sectional change point.  Function of thinning out point cloud data comprises process of creating a filter and process of interpolation of the point cloud data.  Process of creating a filter creates a breakline filter by superimposing breaklines and a grid-like filter.  Next, process of interpolation of point cloud data interpolates the point cloud data using the breakline filter.  This process converts the point cloud data measured at irregular intervals to the thinned-out point cloud data in a grid-like shape with breaklines taken into consideration. 3 FUNCTION OF CANDIDATE BREAKLINE GENERATION The function of candidate breakline generation creates candidate breaklines from the point cloud data.  Here we assume the point cloud data measured by the laser scanner as P={p1, p2, p3, ..., pi}.  This function solves the aforementioned problem of freshness due to the update cycle of DM data. 3.1 Process of dividing elevation This processing divides point cloud data into multiple layers as pre-processing to distinguish the region of the crown surface from the point cloud data.  Specifically, the point cloud data P is divided at regular intervals at every height h, as shown in Figure 2.  The layer set acquired from this processing is assumed as L={l1, l2, l3, ..., lj}. 319-3 XZYXZYhXZYl1l2l3Point cloud data P Divided at regular intervals h Divided layer L  Figure 2: Process of dividing elevation 3.2 Process of clustering density In order to distinguish the layer on which point cloud data are concentrated, this processing performs clustering by the density of points over the layer and calculates the region area of each cluster.  The characteristics of the point cloud data obtained by measuring a river levee with MMS shows that the density of the point cloud data of the crown surface close to the measurement vehicle tends to be high, and that the point cloud data density of those away from the measurement vehicle such as slope or foot of slope tends to be low.  For this reason, this processing used DBSCAN method, which is one of the data mining methods, as a method for clustering based on density.  DBSCAN is a clustering method based on the point density using two values: distance threshold Eps, and threshold of target numbers MinPts.  Figure 3 shows a method of process of clustering density.  First, the point cloud data contained in an arbitrary layer l j, Pdbscan={pd(j, 1), pd(j, 2), pd(j, 3), ..., pd(j, l)}, is clustered by DBSCAN.  When arbitrary point pd(j, l) in the point cloud data Pdbscan and neighborhood point pd(j, l+1) contained in the neighborhood point cloud NEps(pd(j, l)) within the distance Eps from that point satisfy the following conditional equations [1] and [2], the neighborhood point pd(j, l+1) is classified as the same cluster with the point pd(j, l). [1]  )( )1,(),( +∈ ljEpslj pdNpd   [2]  MinPtspdN ljEps ≥+ )( )1,( Assume the clustering result by DBSCAN from an arbitrary layer l j as Clj={cl(j, 1), cl(j, 2), cl(j, 3), ..., cl(j, k)}.  Then obtain the region outline of each cluster cl(j, k) of the cluster set Clj by the active contour method.  In addition, assume the point cloud data applied to the active contour method as Psnake={ps(j, 1), ps(j, 2), ps(j, 3)}.  Region area Sl of the cluster set ClPoint cloud data of Layer ljXYCluster set Cl of layer ljXYXY Cluster cl(j, 1)Area sl(j, 1)Cluster cl(j, 2)Area sl(j, 2)Cluster cl(j, 3)Area sl(j, 3) Figure 3: Process of clustering density Figure 4 shows the outline of the active contour method used in this research.  The active contour method is a method for capturing the contour of an object by repeating movement and deformation of the closed curves that contain the object.  First, create a circle so that it circumscribes the target point cloud data Psnake.  Second, create point-series Ptangent={pt1, pt2, pt3, ..., ptn} arranged at equal intervals on the 319-4 circumference, and a tangent line of the circle T={t1, t2, t3, ..., tn} passing through ptn.  Then, translate the tangent line of the circle tn passing through the point ptn on the circumference towards the center point of the circle until it overlaps with the target point cloud data ps(j, m).  Then the distance translated by ptn,  range is calculated using equation [3].  Here tn is a straight line formula ax+by+c=0.  xt and yt are assumed as x and y coordinates of the translating point ptn, whereas xs and ys as x and y coordinates of the target point cloud data ps(j, m). [3]  22 bacybxacybxaRange sstt++⋅+⋅−+⋅+⋅=  Point-series placed at regular intervals on the circumference PtangentPoint cloud dataXYXYObtaining a set of tangent lines T’ of point cloud dataT’XYTranslating the tangent line tn of the circleXY tn Figure 4: Outline of the active contour method used in this study The set of tangent lines of point cloud data obtained by this processing is assumed as T'={t'1, t'2, t'3, ..., t'n}.  The set of intersections of a tangent line set T' is assumed as the region outline F={f1, f2, f3, ..., fo}.  Then the region area Slj={sl(l, 1), sl(l, 2), sl(l, 3), ..., sl(l, k)} of the cluster set Clj  is calculated from the region outline F of the cluster obtained by the active contour method.  Region area Slj of each cluster is calculated by calculating the area of the following polygons using equation [4]. xo and yo in equation [4] are x and y coordinates of vertex fo of the region outline F.  xo+1 and yo+1 are x and y coordinates of the point fo+1 adjacent to vertex fo. [4] ∑=++ +⋅−=Foooookj yyxxsl111),( )()(21  3.3 Process of identification crown surface Figure 5 shows the process of identification crown surface.  The process of identification crown surface identifies the layer of the crown surface in order to acquire the candidate breakline.  The layer of the crown is estimated using the ratio of the region area of the point cloud data belonging to the layer to the sum total of region area of each cluster Slj, or the region size where there are point cloud data.   Then, the region area of each layer is assumed as S={s1, s2, s3, ..., sj}.  These region areas are different from layer to layer since they depend on the result of clustering processing of each layer. First, when it is a micro cluster of which the region area is threshold MinSize or smaller, the micro cluster is removed from any layer l j.  Next, the region outline and region area of any layer l j is calculated using the active contour method like the preceding term.  Then using the region area of the layer and the region area of the cluster contained in the layer, layer lj that is to be the crown surface is identified using equation [5].  Finally, the region outline of the layer l j of the crown surface is obtained as the candidate breakline BL. 319-5 Table 1: Experiment environment Type Specifications Hardware Experiment equipment CPU Intel® Core™2 Duo CPU 2.50Ghz Memory 4.0GB HDD 280GB Software Developing  environment and language Visual Studio 2010 Visual C# CAD AutoCAD Civil3D 2010  Table 2: Details of the experimental data Items MMS data Number of points About 4.1million Measuring distance 800m Absolute precision Horizontal direction 10cm Vertical direction 15cm Relative precision 1cm  4.3 Experiment description First, three 3D models are generated by entering the point cloud data. • Existing method (Method a) (Tanaka, 2010) • Proposed method with a function of candidate breakline generation added to the existing method (Method b) • Proposed method with a function of candidate breakline generation and process of removal wall-shaped noise added to the existing method (Method c) Next, cross sections are obtained from the generated 3D model.  Using the longitude and latitude of the distance marks in the measured cross sections of correct data, information on the same place is extracted as to cross sections.  Finally, the cross sections is compared with their correct data for evaluating the precision. 4.4 Experimental results and discussion This experiment evaluates the precision using the difference in elevation of each cross section by superposing the cross section of a 3D model created through each method with the measured cross section.  For the measured cross sections to be used in this experiment, 5 drawings were prepared at every 200m between the distance marks of 7.2km and 8.0km points contained within the measurement area using MMS.  Figure 8 shows an image of comparing cross sections.  Here comparison of the difference in elevation is calculated over the evaluation points created by dividing the measured cross section at intervals 1cm.  Moreover, the comparison results are classified into three sections: all, section A and section B, and totalized as shown in Figure 9. Difference in elevationElevation (m)Distance (m)Cross section of the 3D modelMeasured cross section Figure 8: Image of comparing cross sections Overall Section A5cm 5cmSection B Section B  Figure 9: Groups of totalizing comparison results Section A is assumed as the range including the width of the crown surface with 5cm added on both sides.  This is determined with reference to the value of the work progress control standard for filling works of river earthwork providing the accidental error for the crown surface should be within 10cm. Section B is assumed as the range not contained in Section A except the crown surface.  Table 3 shows the totalized results for 3D models created using each method and the measured cross section drawing.  319-7 This experiment totalized to what degree the difference in elevation between cross sections is contained within 15cm at intervals of 5cm.  This is a reason there is absolute error of 15cm (in the vertical direction) in the measurement data using MMS.  Furthermore, for each totalized section, underlines were added to the values with the lowest error between the cross sections created using three methods and the measured cross section drawings.  The experimental results proved the following three points. Table 3: Comparison results between 3D models and measured cross sections Measured points Overall (%) Section A (%) Section B (%) Method  a Method  b Method  c Method  a Method  b Method  c Method  a Method  b Method  c Distance mark  7.2km ~ 5cm 22.68  20.45  23.64  64.86  81.08  91.89  17.03  12.32  14.49  ~10cm 29.71  29.07  37.38  72.97  100.00  100.00  23.91  19.57  28.99  ~15cm 35.46  33.23  45.37  78.38  100.00  100.00  29.71  24.28  38.04  Distance mark  7.4km ~ 5cm 6.33  10.67  8.33  0.00  0.00  0.00  6.99  11.76  9.19  ~10cm 16.00  25.33  21.67  0.00  0.00  3.57  17.65  27.94  23.53  ~15cm 29.00  41.33  49.33  60.71  7.14  75.00  31.99  44.85  46.69  Distance mark  7.6km ~ 5cm 2.33  2.33  2.84  0.00  0.00  2.44  2.69  2.69  2.99  ~10cm 5.43  5.43  15.50  0.00  0.00  90.24  6.29  6.29  6.89  ~15cm 11.37  8.79  20.67  12.20  0.00  100.00  11.68  10.18  11.68  Distance mark  7.8km ~ 5cm 23.49  16.95  24.94  53.06  0.00  73.47  19.51  19.23  18.41  ~10cm 37.29  37.77  37.77  81.63  81.63  91.84  31.32  31.87  30.49  ~15cm 50.36  50.85  49.15  100.00  100.00  100.00  43.68  44.23  42.31  Distance mark  8.0km ~ 5cm 15.38  15.38  21.47  48.94  59.57  85.11  9.43  7.55  10.19  ~10cm 27.88  27.56  33.65  89.36  100.00  100.00  16.98  14.72  21.89  ~15cm 53.21  50.96  54.49  100.00  100.00  100.00  44.91  42.26  46.42  Overall average ~ 5cm 14.04  13.15  16.25  33.37  28.13  50.58  11.13  10.71  11.05  ~10cm 23.26  25.03  29.20  48.79  56.33  77.13  19.23  20.08  22.36  ~15cm 35.88  37.03  43.80  70.26  61.43  95.00  32.39  33.16  37.03  First, the total average in Table 3 reveals that the proposed method c realized the smallest errors compared with the measured cross sections throughout the overall part, Section A, and Section B.  The average of the total results for Section A reveals that about 95% is included within 15cm, which is the allowable error for MMS, indicating that a 3D model with high precision has been generated.  This is a reason the vegetation noise in the upper part of the crown surface is properly removed. Second, Section A with the distance mark of 7.6km in Table 3 reveals that the 3D model generated by Method c is able to reproduce the crown surface more precisely than the ones created by the other two methods.  Figure 10 shows the result of superposing the cross section at the distance mark of 7.6km with cross section drawings created from the 3D models by each method.  Figure 11 reveals that the cross sections of 3D models created by Methods a and b do not agree with the measured cross sections in the part of crown surface, indicating that the 3D models are generated at higher position than the measured cross section.  On the other hand, the 3D model created by Method c is able to have represented the present shape, showing the similar shape of the crown surface as that of the measured cross section.  When analyzing the point cloud data at the distance mark 7.6km, there were wall-shaped noises at the upper part of crown surface including the top of slope.  On the other hand, the 3D model generated using the proposed method was able to remove the wall-shaped noise on the crown surface with high precision. 319-8 In this research, all the point cloud existing on the crown surface was assessed as noise in order to estimate the boundary lines of vague planes represented by point cloud data.  In the future, demonstration experiments will be conducted to verify to what degree of differences occur between the 3D model generated using such a noise removal method and actual present topographic features, and the applicability to practices and effects to be enjoyed will be made clear. This research generated 3D models for a river levee, and since the proposed method of this researchhas high versatility, its applicability to other civil engineering structures (roads, tunnels, or dams) will be examined in the future.  Moreover, though the applicability was verified with focus on construction stages (information-oriented construction), some of the challenges to solve in future can be development into a method for assisting creation of register maps for maintenance as well as a management approach using spatial-temporal model added with time base. Finally, this paper is part of the outcome of the activities of "Workshop on establishing a data circulation environment for river projects" set up in Kinki Regional Development Bureau, MLIT for the purpose of promoting new technology development with public and academic sectors as its core.  We plan to make further studies to realize environment for smooth data circulation. Acknowledgements We would like to thank the members at MLIT Kinki Regional Development Bureau MILT for their Invaluable advice given on this paper, Nippon Koei Co.,Ltd.  and Mitsubishi Electric Corporation for their cooperation in measurement of the point cloud data used in this research. References Tanaka, S., Imai, R. and Nakamura, K. : Examination of Framework for the Development of Data Distribution Environment for Public Works, Journal of Applied Computing in Civil Engineering, Japan Society of Civil Engineers, Vol.18, pp.37-46, 2009. (In Japanese) Fukumori, H., Sada, T. and Okubo, H. : Study on the Measurement Accuracy of 3D Laser scanner, Journal of Applied Computing in Civil Engineering, Japan Society of Civil Engineers, Vol.18, pp.193-200, 2009. (In Japanese) Ohtu, S. and Sada, T. : Method of Processing Data with a Cloud of Points on the 3-D Shape Surveying, Journal of Applied Computing in Civil Engineering, Japan Society of Civil Engineers, Vol.16, pp.27-36, 2007. (In Japanese) Taniguchi, T. and Mashita, K. : Generation of 3-Dimensional Domain from Point Clouds and Its Finite Element Modeling, Bulletin of the Japan Society for Industrial and Applied Mathematics, The Japan Society for Industrial and Applied Mathematics, Vol.15, pp.310-319, 2005. (In Japanese) Shaohui, S. and Carl, S. : Aerial 3D Building Detection and Modeling From Airborne LiDAR Point Clouds, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, IEEE, Vol. 6, No.3, pp.1440-1449, 2013. Choi, Y., Jang, Y., Lee, H. and Cho, G. : Three-Dimensional LiDAR Data Classifying to Extract Road Point in Urban Area, IEEE Geosciences and Remote Sensing Letters, IEEE, Vol.5, No.4, pp.725-729, 2008. Tanaka, S., Imai, R., Nakamura, K. and Kawano, K. :Research on Generat       Breakline from Point Cloud Data, Proceedings of the 10th International Conference on Construction Applications of Virtual Reality, CONVR2010, pp.347-356, 2010. 319-10  RESEARCH FOR GENERATING 3D MODEL FROM LASER SCANNER DATA REMOVED NOISEShigenori Tanaka 1, Ryuichi Imai 2, Kenji Nakamura 3, Kouhei Kawano4, and Satoshi Kubota 51 Faculty of Informatics, Kansai University, Japan2 Faculty of Engineering, Tokyo City University, Japan3 Faculty of Information Technology and Social Sciences, Osaka University of Economics, Japan4 Graduate School of Informatics, Kansai University, Japan5 Faculty of Information Environmental and Urban Engineering, Kansai University, Japan1Subject• In Japan, the Ministry of Land, Infrastructure, Transport and Tourism has been taking the lead in promoting measures for computerization.– When focusing on the construction phase within the life cycle of a public work, computer-aided construction, or Intelligent Construction, is propelled eagerly.• In intelligent construction, three-dimensional (3D) CAD data (or 3D Model) that represent construction objects or construction site terrain are often utilized.– Consequently, technologies to create the 3D CAD data of construction sites efficiently with high precision are required.2Generating 3D-CAD Data PlaneCross-SectionLengthwise-SectionAssembling 3D-CAD data from 2D-data3Lifecycle of Construction InformationDigital mapping data Modifying structure“Survey data” Estimation人工衛星Intelligent constructionSystem of work progress controlUpdating data to GIS“Construction data”“Design data”3D data4Point Cloud Data for 3D-CAD Using New Media5Tool and Machine for Surveying Disaster • Mobile Mapping System(MMS)– Each method gets many point cloud data.• Laser Profiler (LP)Surveying by LP(http://www.nilim.go.jp/lab/rcg/newhp/seika.files/lp/ )Surveying by MMS(Mitsubishi Company,http://www.mitsubishielectric.co.jp/pas/mms/ )6- Case of Embankment of River using MMS-7Point Cloud Data of Embankment of RiverPoint Cloud Data of Embankment3D-CAD Dataautomatic generation 8Point Cloud Data of Embankment of River9Point Cloud Data of Embankment of RiverBreak line as alignment10Point Cloud Data of Embankment of Riverbeforeafter11River embankments have convex shapeOur Approach-10-50510-100 0 100 200 300 400 500 600 70012Variations in density depending on the distance from MMSOur Approachvehicular swept path of MMS13Processing FlowFunction of candidate breakline generationFunction of breaklinesextraction BreaklinedataThinned-out point cloud data(3D CAD input data)OutputFunction of thinning out point cloud dataOutputPoint cloud dataInput Candidate breaklineInputOutputInput14Input DataMost of vegetation noise is found on the crown surface and top of slope.Artifact noise on the crown surface included passersby and passing vehicles15Processing flowFunction of candidate breakline generationFunction of breaklinesextraction BreaklinedataThinned-out point cloud data(3D CAD input data)OutputFunction of thinning out point cloud dataOutputPoint cloud dataInput Candidate breaklineInputOutputInput16BACProcess of dividing elevationXZYABCeXYBC17BACProcess of dividing elevationAXZ BXZ CXZXZYBCB18Process of clustering densityRegion area of the cluster set CAPoint cloud data of Layer AXZCluster sets CAof layer AXZXZ Cluster1Cluster319BACProcess of identification crown surfaceAXZ BXZ CXZXZYBCBArea of Cluster1Area of Cluster1Area of Cluster2Area of Cluster1Area of Cluster2Total Area Total Area Area20Process of identification crown surfacePercentage=∑Cluster Areas ÷ Total AreaPercentage= 95% Percentage = 70% Percentage = 50%21AXZ BXZ CXZBArea of Cluster1Area of Cluster1Area of Cluster2Area of Cluster1Area of Cluster2Total Area Total Area Area21050010001500200025001 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17Process of identification crown surfacePercentageLayer NoAXZCluster1AreaPercentage = 95%identification crown surface22Process of identification crown surfaceAXZCluster1Areacrown surface 23Processing flowFunction of candidate breakline generationFunction of breaklinesextraction BreaklinedataThinned-out point cloud data(3D CAD input data)OutputFunction of thinning out point cloud dataOutputPoint cloud dataInput Candidate breaklineInputOutputInput24Wall-shaped noise removal processcrown surface removes the point cloud data with the altitude higher than the crown surface 断面モデルcross section model XYXZYXZY25Cross-sectional change point identification processingThreshold β in identifying a horizontal lineIdentifying the cross-section change pointchange pointcross section model Searching for horizontal lines from the cross section model threshold β XYXYXYcandidate breakline26Process of creating breaklines断面変化点・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・●A●C●E●B●D断面変 点ZXcross-sectional change points ブレイクライン・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・●A●C●E●B●Dブレ インZXbreaklines・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・●A●C●E●B●D断面変化点集合断面変 合ZXsmoothing with using the median filter 27Processing flowFunction of candidate breakline generationFunction of breaklinesextraction BreaklinedataThinned-out point cloud data(3D CAD input data)OutputFunction of thinning out point cloud dataOutputPoint cloud dataInput Candidate breaklineInputOutputInput28Function of thinning out point cloud data・ ・ ・・ ・ ・・ ・・ ・ ・・ ・ ・・ ・ ・・ ・ ・ ・・ ・・ ・ ・ ・・ ・・ ・ ・ ・・ ・・ ・ ・ ・・ ・・ ・ ・ ・・ ・・ ・ ・ ・・ ・・ ・ ・・ ・ ・・ ・・ ・ ・・ ・ ・・ ・ ・・ ・ ・ ・・ ・・・・・・・・・・・・・・grid-like filterbreaklinebreakline filterpoint cloud data thinned-out point cloud data interpolates point cloud data   29- EVALUATION EXPERIMENT -30experimental dataYodo RiverMeasurement range7.2 km8.0 km31Result of visualizing MMS point cloud dataItems MMS dataNumber of points About 4.1millionMeasuring distance 800mAbsolute precision Horizontal direction 10cmVertical direction 15cmRelative precision 1cm 32Experiment description(1/2)• First, three 3D models are generated by entering the point cloud data– Method a• Existing method (Tanaka, 2010)– Method b • Proposed method with a function of candidate breakline generation added to the existing method – Method c• Proposed method with a function of candidate breakline generation and process of removal wall-shaped noise added to the existing method 33Experiment description(2/2)• Next, cross sections are obtained from the generated 3D model.• Finally, the cross sections is compared with their correct data for evaluating the precision.correct data(the measured cross section)34Evaluation Method(1/2)Cross-Section Model(created from 3D Model)Correct data(measured cross section) overlay1cmExpansionthe difference in elevation of each cross section35Evaluation Method(2/2)Overall Section A5cm 5cmSection B Section B 36Experimental results(1/2)Average 3D ModelMethod A Method B Method CSection A (%)~ 5cm 33.37 28.13 50.58 ~10cm 48.79 56.33 77.13 ~15cm 70.26 61.43 95.00 Section B (%)~ 5cm 11.13 10.71 11.05 ~10cm 19.23 20.08 22.36 ~15cm 32.39 33.16 37.03 Overall (%)~ 5cm 14.04 13.15 16.25 ~10cm 23.26 25.03 29.20 ~15cm 35.88 37.03 43.80 37Experimental results(2/2)• 3D model generated from point cloud dataMethod b Method c38Conclusion• This research developed the method of automatically generation of a 3D model.• Specifically, technologies for handling the point cloud data were developed.– estimate the crown surface of a river levee– remove wall-shaped noise contained in point cloud data 39FinishThank you for your attention !40Experimental results3456789-40 -30 -20 -10 0 10堤防の高さ(m)堤防の幅(m)3456789-40 -30 -20 -10 0 10堤防の高さ(m)堤防の幅(m) Method b Method c   Width of levee (m) Width of levee (m)Height of levee (m)Height of levee (m)   Legend: Cross section created from the 3D modelMeasured cross section at the distance mark of 7.6km41

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.52660.1-0076443/manifest

Comment

Related Items