A traffic incident detection model for arterial streets equipped with fixed and mobile detection systems

Intelligent Vehicle Highway Systems have created new traffic monitoring approaches, and fueled new interest in Automated Incident Detection (AID) Systems. One new monitoring approach utilizes actual travel times experienced by vehicles, called probes, equipped to transmit this information to a control center. The database needed to calibrate Arterial AID Systems based on probe and detector reports is nonexistent. INTRAS, a microscopic traffic simulation package, is selected and enhanced to generate incident and incident-free probe and detector reports on surface streets. The detector and probe data generated follow expected patterns in their departure from normalcy under incident conditions. A novel approach to incident detection on arterial that utilizes multi-class and multivariate classifiers to differentiate between various traffic states is proposed. The possibilities for incident detection within the low to moderate flow range seem promising, given the observed model sensitivity to lane flow imbalance. The similarities between the Bayes' fusion of multi-sensor allocations and Multiple Attribute Decision Making (MADM) is established. An array of MADM fusion algorithms is thus made available to the traffic engineer. The probe data proves valuable in enhancing the performance of detector data based models. Models based solely on probe data lack in performance due to excessive overlaps in class distributions.