Image Sequences Based Traffic Incident Detection for Signaled Intersections Using HMM

For an intelligent transportation system (ITS), traffic incident detection is one of the most important issues, especially for urban area which is full of signaled intersections. In this paper, we propose a novel traffic incident detection method based on the image signal processing and hidden Markov model (HMM) classifier. First, a traffic surveillance system was set up at a typical intersection of China, traffic videos were recorded and image sequences were extracted for image database forming. Second, compressed features were generated through several image processing steps, image difference with FFT was used to improve the recognition rate. Finally, HMM was used for classification of traffic signal logics (East-West, West-East, South-North, North-South) and accident of crash, the total correct rate is 74% and incident recognition rate is 84%. We believe, with more types of incident adding to the database, our detection algorithm could serve well for the traffic surveillance system.