Deployment of Intelligent Transportation Systems (ITS) falling under the categories of advanced traffic management systems (ATMS) could considerably reduce incident response/clearing times, traffic delay and minimize congestion. Successful ATMS rely on timely detection of any traffic disruptions or occurrences of incidents. As such, one of the key components of an ATMS is the traffic incident detection system.
Through the use of real-time traffic data collected by a vehicle detection system (VDS), incident detection systems serve to alert operators in the Traffic Management Center (TMC) to potential incidents to be confirmed via CCTV traffic cameras. As fundamental as these systems were to daily operation, the performance of these systems was far from perfect. Primarily, the problem stemmed from the fact that different traffic, geometric and environmental conditions at different locations require different detection algorithm parameters for each detector station. As the aforementioned conditions changed, so too should have the algorithm parameters.
In addition, typically, incident detection systems used loop detectors in order to collect real-time traffic volumes, speeds and occupancy. Loop detectors deteriorated with time and were often torn up during construction. With each new installation, the loop detector’s sensitivity varied requiring re-calibration of the incident detection algorithm. Maintaining the overall system (e.g. constant reinstatement of loop detectors, re-calibration of the system, etc.) for optimal performance became both time consuming and costly.
Research and development of incident detection algorithms and systems were traditionally done offline where the focus was to perfect the algorithm or methodology while maintaining the assumption of a perfect data set. Exposure of these incident detection systems to real-world traffic data potentially revealed opportunities for significant improvement and identify initiatives for further research. However, while researchers and practitioners alike recognized the benefits to testing and developing these algorithms in an “online” environment, there was the concern that such efforts may intrude on the daily operation of the traffic management centers.
The University of Toronto ITS Centre and Testbed provides a unique opportunity for testing, evaluation and development of traffic incident detection systems. Bearing the resemblance of a typical traffic control centre, the ITS Centre has access to real-time streaming traffic data feeds as collected by the existing loop detectors from Ministry of Transportation, Ontario (COMPASS) and City of Toronto (RESCU). In addition, the ITS Centre features a 20 monitor video wall with access to all of the COMPASS and RESCU video feeds through 20 simultaneous video channels. As such, the centre can facilitate research, testing and development in a manner that is non-intrusive to the daily operations of RESCU and COMPASS.
The purpose of this project was to evaluate innovative automated incident detection systems and to explore new possibilities with respect to traffic probe data sources.
A key component to this project lied in the development of an Incident Detection Testbed System (IDTS) that uses real-time, online video and data feeds to evaluate incident detection systems. It was anticipated that the real-time, online evaluation of these systems would yield: a clear distinction for the strengths and weaknesses of the respective systems, recommendations for more effective deployments of these systems, and potentially, recommendations for new areas of research to further improve the operation of incident detection systems. Within the scope of this assignment, three incident detection systems were evaluated and compared using the IDTS: the McMaster Algorithm, the University of Toronto Genetic Adaptive Incident Detection GAID, and the Citilog Video Vehicle Detection System (VVDS).
Concurrent with the evaluation of present-day incident detection systems, an investigation was conducted into the potential for using probe-based data (e.g. traffic data extracted from transponder systems, cell phones, electronic license plates, etc.) for the purposes of traffic management. This may well represent the future of traffic management systems as the costly requirement for vehicle detection infrastructure expansion, re-instatement and maintenance continues to grow. The investigation yielded a report identifying the issues surrounding the potential migration from traditional point-detector based data to probe-based data.
The overall project focused on exploring the “real world” issues associated with deployment of incident detection systems and the potential use of probe based data for traffic management purposes. A final report and executive summary was provided that clearly identified the issues, quantify them, and propose realistic, cost-effective solutions wherever possible.
In the context of this project, the following deliverables were also anticipated:
- A set of usage guidelines for current incident detection system deployment
- Identification of new measures of effectiveness (MOEs) for incident detection system evaluation
- Recommendations for further research in the field of incident detection algorithms and potentially probe-based traffic management algorithms