Semantic Hierarchy Based Reasoning Chain Algorithm for Event Detection on an Intersection

In this paper, a method of vision-based automatic incident detection for traffic surveillance systems at crossroad is presented. We have developed a dedicated vehicle tracking algorithm based on the ST-MRF model (S. Kamijo and M. Sakauchi, 2002), and have done a successful work of incident detection for an high-way monitoring system (M. Harada et al., 2004) using the tracking algorithm. Here, we apply incident detection to a crossroad. Since the traffic situation on crossroad is much more complex than high-way road, an efficient method to classify the various vehicles\´ behaviors is required. Even though there is no white lines specify the routes for vehicles to drive in the central area of a crossroad, the vehicles will still follow some unseen routes. Here we call these unseen routes "semantic routes". Our detection method detects the semantic routes and manages vehicles driving in the same semantic route as a "vehicle reasoning chain", and mainly focuses on the first vehicle of a chain. This method can find an incident vehicle quickly in a heavy traffic situation with less false detects