Vehicle tracking by fusing multiple cues in structured environments using particle filter

In this paper, we present an effective and robust visual vehicle tracking algorithm using particle filter and multiple cues. A stable histogram-based framework is extended to evaluate color, edge, texture and motion cues in structured environments. This framework is suitable for practical conditions since in many applications the object motions are limited by structure of the surveillance scene. We show the appropriate method to model the likelihood function of each cue. However motion cue is irregular, so generating the corresponding distribution from its likelihood function and using the structure of environment as likelihood decision function can handle this problem. For modeling the environment, distance transform is used. In addition, noise parameters and the fusing weight of cues are obtained adaptively. Experimental results on several video surveillance sequences show the effectiveness and robustness of proposed method.