Techniques for Detection of Incidents and Traffic Disturbances

The increasing contribution of incidents to freeway congestion has generated strong interest in the development of incident detection algorithms in the last two decades. According to Federal Highway Administration estimates (Lindley, 1986), incidents currently account for up to 60% of the vehicle-hours lost to freeway congestion; projection for the year 2005 indicates a 70% contribution of incidents to total delay. Fast and accurate detection of incidents can, therefore, substantially reduce the impact of incident congestion on freeway traffic. In particular, when an incident alarm is promptly signaled, traffic management plans can be adjusted in real time to produce the best control and guidance actions in freeway corridors. In addition, the incident management process (detection, response, and clearance) is initiated as emergency vehicles can be promptly dispatched to clear the incident. Existing techniques for the detection of freeway incidents do not provide the necessary reliability for freeway operations. Conventional automated techniques, based on computerized algorithms, are less effective than is desirable for operational use because they generate a high level of false alarms. Operator-assisted methods minimize the false alarm risk, but suffer from missed or delayed detections, are labor intensive, and restrict the potential benefits from advanced, integrated traffic management schemes. The initial phase of this research focused in assessing the performance limitations of conventional automatic incident detection systems. That research was directed towards two objectives, the performance evaluation of major existing algorithms and the development of an improved algorithm. This part of the research pointed out that the existing techniques for the automatic detection of freeway incidents are not reliable as they are seriously handicapped by excessive, operationally unacceptable false alarm rates. The new algorithm proposed by the authors was developed for identifying capacityreducing incidents in freeway traffic. That algorithm aims to minimize the number of false alarms that the existing algorithms generate when temporal random oscillations in the traffic measurements, frequently observed in congested flows, occur. The proposed structure involved preprocessing the traffic data with average, median, or exponential smoothers over data windows of approximately five minute length to eliminate or reduce the size of traffic fluctuations. Although the new algorithm showed an improved and satisfactory performance relative to the conventional algorithms, the initial stage of this research pointed out the need of more research in finding ways and methods for distinguishing between the incident and the non-incident alarms and highlighted the issues that had to be addressed by the second stage of this project.