Abstract
Vehicle recognition is a nontrivial problem in Vision-based Intelligent Traffic System (ITS), especially when the occlusion occurs among the vehicles. In this thesis, we propose a method to segment moving vehicle from scene. Since the speed of the vehicles is different, the pixels of vehicle objects also have different motion vectors in consecutive frames. We cluster the different motion vectors to separate the vehicles in occlusion condition. Our system consists of background subtraction, object segmentation and vehicle classification. In the background subtraction, we develop an adaptive background model to extract foreground object, and followed by noise remove, closing, opening, and l ling. Since the foreground objects may occlude each other we have developed a method to divide the occluded object into two different objects. Finally we classify the segmented vehicles