ARTERIAL INCIDENT DETECTION USING FIXED DETECTOR AND PROBE VEHICLE DATA

This paper describes incident detection algorithms for urban arterial streets using two distinct data sources: fixed traffic detectors and p robe vehicles. The data sources are used independently to obtain two d istinct algorithms. This approach is undertaken to increase the overal l coverage of incident detection capabilities as early implementation will result in relatively few cases when data is available from both f ixed detectors and probe vehicles on the same link and during the same time period. The algorithms were developed using simulation data for the ADVANCE ITS operational test; they will subsequently be recalibrat ed with field data collected during the ADVANCE demonstration project. Discriminant analysis was used to estimate a variety of models based on different traffic flow measures from each data source. Various func tions of fixed detector measures (volume and occupancy) and probe vehi cle travel times were considered for inclusion in the fixed detector a nd probe vehicle algorithms, respectively. The most effective variable s for detecting incidents were volume divided by occupancy (which is r elated to average speed) for fixed detectors and average speed for pro be vehicles. In both cases, traffic measures for the incident link wer e most useful for incidents located in the downstream portion of the l ink and for the next upstream link for incidents located at the upstre am end or in the middle portion of the link. Further, it was generally found that data from a single link provided almost equally good incid ent detection as data obtained from pairs of links. This led to the de velopment of an algorithm that uses data from a single detector or lin k, thereby supporting incident detection on any link that has a curren t data independent of data availability from other links. The performa nce of the algorithms was evaluated using detection rates and false al arm rates, which were found to be in the same range for both the algor ithms. The fixed detector algorithm showed better detection ability, b ut its use is limited by the number of detectorized links in the netwo rk, while the performance of the probe vehicle algorithm was dependent on the number of reports available per time period.