Multi-state and multi-sensor incident detection systems for arterial streets

Incident detection systems typically emphasize incident presence and location over incident severity and incident recovery. Yet, Advanced Traveller Information Systems (ATIS) and Advanced Traffic Management Systems (ATMS) rely on the latter states to implement and terminate diversion, and its supportive control strategies. Further, incident detection systems directly benefit from processing measurement vectors rather than scalars. Vectors of lane measurements favor detection through lane imbalances and identification of incident host lanes. Intelligent Transportation Systems promise new sensor data to control centers, including the travel times experienced by probe vehicles. Vectors of new and old sensor inputs may possess enhanced discriminatory values. To accommodate added detection states and the fusion of multi-sensor input vectors, this paper reformulates the arterial incident detection problem as a multiple attribute decision making problem with Bayesian scores. This novel approach utilizes as input the combinations of simulated probe travel times, number of probe reports, lane specific detector occupancies and vehicle counts. Model based solely on probe data lack in performance due to excessive overlaps in class distributions. Models based on detector occupancies and vehicle counts by lane perform outstandingly. They display a propensity to detect through lane measurement imbalances. The probe data are shown to enhance the performance of detector data based models.