Scene model for video traffic analysis
Mathieu Brulin, Henri Nicolas · 2014
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.