Scene model for video traffic analysis

For traffic video surveillance applications, it is necessary to define automatic systems able to analyze vehicles or pedestrians behaviors in real time. This is a very difficult task which involves major difficulties such as moving objects detection, tracking, and classification. In order to simplify it, a priori knowledge of the scene characteristics can be exploited. This paper presents a complete system able to automatically analyze the scene geometry and the traffic rules in the context of a traffic scene analysis. It allows detecting the road area, its decomposition into circulation lanes and forbidden areas, the typical vehicle´s trajectories and some circulation rules. This scene model can be usefully exploited by a video traffic analysis system. Experiments show that the scene model is well estimated.