Vehicle detection and recognition are two important issues in vision-based traffic applications. We are more interested in recognizing than detecting moving vehicles in real road scenes captured by a static camera. We first extract regions corresponding to moving vehicles in the image by an image difference-based method using the adaptive Kalman filtering. But it is difficult to recognize vehicles n real time using the previously proposed algorithms. In this thesis, we proposed a fast vehicle recognition algorithm in real road scenes captured by a static camera.
We first extract regions corresponding to moving vehicles in the image by an image difference-based method using the adaptive Kalman filtering. We recognize vehicles by analyzing edge distributions of these regions. In this thesis, we recognize five kinds of vehicles- sedan, truck, bus, jeep and van. In real road scenes, it is difficult to obtain lines and significant features. Therefore we use the models of edge clusters to recognize vehicles. We obtain dominant edge clusters, and find matched objects in the scene based on the relative distance between these edge cluster. Although the proposed method has some limitations such that it assumes fixed camera and that vehicles must keep their direction, it is robust in noisy scenes and is fast due to its simplicity. The proposal algorithm shows the possibility of real-time recognition.