Vehicle Modeling and Detection in Crossroads

In this work, we proposed a novel method to detect vehicle automatically under different field of view in urban area. First, traffic flow maps are reconstructed to estimate the size and orientation of a vehicle for every position. Vehicle models are then represented by flow information and foreground blob which shows the potential appearance of a vehicle in each position in the crossroad. Finally, we perform a voting strategy to detect the potential vehicle. The experimental results show the robustness of our system. Our algorithm can be applied to the surveillance cameras with different view in crossroads.