Incident Detection on an Arterial Roadway

Presented here is the development of an automatic incident detection algorithm for use on Lakeshore Boulevard, Toronto, Canada, based on volume or occupancy data recorded from fixed-loop detectors. Four prospective logics were based on 20-sec intervals; the remaining five were based on traffic-signal cycle lengths to eliminate the fluctuations in 20-sec data. To identify the detection ability of each logic, data from a known severe incident (a two-lane blockage) were used. Only one logic exhibited promising results from the initial development and feasibility test: the logic that compared current cycle volume and occupancy values with those averaged over the previous 3-, 5-, and 10-cycle periods. Further evaluation was conducted on this logic. A data base was developed around two additional reported serious incidents; the logic detected both incidents before their official start times. A second data base, which consisted of data from 13 days, was developed to test the overall performance of this logic, focusing specifically on false alarms. On average, one unexplainable false alarm was reported for every 5 hr of data tested for the entire Lakeshore system (49 detector stations). Testing with additional Lakeshore incident data in a real-time environment is required to fully investigate the ability of this algorithm to detect serious incidents. As developed, this logic cannot distinguish between congestion due to incidents and recurrent congestion. The latter was responsible for the majority of the alarms as tested and thus future improvements should focus on this aspect.