Vehicle Tracking from Videos Based on Mean Shift Algorithm
Changzhen Xiong, Yi-gui Pang, Zhengxi Li, Yali Liu, Yinghong Li · 2009
Vehicle tracking is one of the essential tasks in a video-based Intelligent Transportation System. A novel method for vehicle tracking is proposed in this paper. First, we analyze the disadvantages of the traditional parameterized model and compare them to the merits of the nonparametric model. Then the mean shift algorithm is deduced by using kernel density estimation in multi-dimensional space. Its convergence is proven using a simple method. In order to overcome the disadvantages of the original mean shift algorithm, this paper proposes an adaptive template and a kernel weights selecting method. The new method accelerates the convergence speed of the algorithm. When we apply this adaptive method, the experimental results show that the proposed method performs well with video-based vehicle tracing.