Incident detection based on short-term travel time forecasting

Prediction of short-term future traffic condition is an important element for route guidance and incident management systems. In this paper, a solution algorithm is proposed for short-term travel time forecasting in congested urban roads of Hong Kong. The travel times in the next 5-minute interval are predicted by using the historical travel time estimates together with their updated temporal variance-covariance relationships. The territory-wide historical travel time database is generated by the real-time travel information system (RTIS) using the automatic vehicle identification data available in Hong Kong. Based on the travel time forecasts and the RTIS travel time estimates, traffic incident can be detected by comparing their differences on the road section before and after the incident. Case studies are presented to evaluate the performance of the proposed algorithms for short-term travel time prediction and incident detection.