Open Collections

Development of a new automatic incident detection system for freeways using a bi-classifier approach

University of British Columbia

As high as 60 to 70% of the traffic delay experienced by motorists in North America is attributed to traffic incidents. Much of this delay is caused by vehicle accidents, vehicle stalls, and other obstructions. A substantial reduction in delay can be achieved by early detection of the incidents that cause it and a prompt response to divert the traffic in the upstream flow. Since the late 60s, Automatic Incident Detection (AID) systems have been developed and implemented to help traffic management authorities. However, high false alarm rates and/or poor performance of the adopted AID systems have caused some authorities to abandon them. The research presented in this thesis discusses the development and assessment of a new AID system. Often after the occurrence of an incident, its "news" travels upstream and downstream through the traffic by means of two waves. Because of some practical difficulties, the information carried by the wave traveling downstream is overlooked by most researchers in this area. In this thesis, it is proposed that through an effective use of the information carried by this wave, it is possible to significantly improve the performance of an AID system. The proposed UBC AID system exploits the information carried by each wave independently and overcomes many of the practical difficulties by adopting a new and unique architecture. The designed architecture not only demonstrates a better performance but also has the ability to maintain the performance over a wide range of operating conditions. Geometric and operational data from a stretch of the Trans-Canada Highway was used to develop a simulation model. This provided a very large set of simulated data under both incident and incident-free situations. Furthermore, it provided an opportunity to examine the performance and robustness of the AID systems over a wide range of geometric and operational conditions. The comparison of the UBC AID method with two other existing and "in-use" systems showed that it is possible to reduce the detection time by about 40% while staying within the desired range of false alarm rates. It was also possible to increase the number of incidents detected within the first few minutes after their occurrence by as much as 2-3 fold.

Thesis/Dissertation

application/pdf

2009-05-28

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/8433

Mechanical Engineering

University of British Columbia

UBCV

DSpace