Improved video-based vehicle detection methodology

Focusing on the problem that the detection accuracy of traffic detection system is sensitive to the changes of complex environments, this paper presents an improved method of vehicle detection. It builds and updates the background adaptively. Additionally, to improve the computation efficiency of shadow elimination, a fast algorithm of neighbor mean based on HSV model is proposed. As the occlusion is inevitable, a new solution is presented to deal with occlusion in this paper. First, a method based on Kalman filter is applied for occlusion identification. And then a search algorithm of template matching based on hierarchical pyramid is utilized for real-time segmentation. Experimental results have shown that the proposed method is effective and high real-time, and it can effectively improve the detection rate of video-based traffic detection system.