AN ANALYSIS OF AN INCIDENT DETECTION METHOD USING TRAFFIC DETECTORS: A TIME-SERIES APPROACH

This paper describes how in recent years the provision of real-time information has become one of the more important means of major traffic management. The VICS (Vehicle Information and Communication System) has been in operation in Japan since 1996; it provides drivers with real-time traffic information, such as road congestion, travel time, parking availability, etc. to local agencies. However, the accuracy of traffic information decreases dramatically, immediately after a traffic incident occurs. It generally takes time to detect an incident, so drivers will encounter congestion caused by recent accidents during their journeys. Therefore, traffic incidents need to be identified as quickly as possible to provide credible traffic information to drivers. An incident-detection system can also be used to effectively prevent secondary accidents from occurring when cars are approaching the site of an accident. With recent progress in information technology, traffic incident detection systems using video cameras and image processing technology have been developed. However, due to budget restrictions and technical difficulties, such systems are installed only at places with a high incident probability. Accordingly, a system that can detect incidents easily and quickly is still required. This paper is aimed at developing a methodology to identify incidents using data obtained from traffic detectors. Time series data could be used for this task, since traffic detectors have already been densely installed along roads. When an incident occurs, there is a change in the traffic flow state. If a time series model can describe the normal traffic flow state sufficiently well, then traffic incidents could be identified from differences between observed and forecast time series data. The contents of this paper are as follows. First, past studies are briefly reviewed and a method for incident detection using a time series model is presented. Second, the characteristic relationship between the volume and time occupancy ratio observed on the Hanshin Expressway in Osaka during traffic congestion is demonstrated. Third, the data processing methods used to check the stationary condition and to transform the stationary data in a time series analysis are shown. Fourth, a time series model for detecting incidents is examined using data from the Hanshin Expressway. Finally, concluding remarks and future studies are presented.