Sampling Strategies for Probe-Vehicle-Based Freeway Incident Detection Algorithms

Different sampling strategies are investigated in terms of percentage of vehicles sampled as probes and three methods of aggregating probe-vehicle data for freeway incident detection. The incident detection algorithm is based on the statistical comparison of the average section travel times of two different groups of probe vehicles. A microscopic traffic simulation model was employed to simulate incidents and collect section travel time data from probe vehicles for evaluation of the sampling strategies. Five probe-vehicle percentages were investigated. For each, the probe-vehicle section travel times were aggregated using three different methods: fixed sample size (FSS), fixed time interval (FTI), and rolling interval. Incident detection performance was analyzed in terms of detection rate (DR), false alarm rate (FAR), mean time to detection (MTTD), number of algorithm applications, and number of false alarms. It was found that the FTI data aggregation method performed best for all the indicators when the probe-vehicle percentage was less than 20. When the probe-vehicle percentage exceeds 30 and both FSS and FTI data aggregation methods have high DRs, the FSS method gave the lowest FAR and the FTI data aggregation method produced the lowest number of false alarm cases and fastest mean time to detection. All three data aggregation methods showed similar performance when the probe-vehicle percentage ranged between 20 and 30.