摘要：In this paper, we propose a super-pixels clique based tracking algorithm to perform track task, meanwhile it can robustly handle the appearance change of target object and heavy occlusion. By two stage adaptive appearance modeling method, we propose the method of learning the target-background appearance framework ,which is based on super pixels principle histogram bins cluster method. The process of computing superpixels cliques confidence not only store the location information of the superpixels, the superpixels cliques recent history and last history also are equally weighted. The first phase of two-stage adaptive cliques constructs and update algorithm is target template superpixels cliques construct stage. By calculating feature distance between superpixels and cliques center, it is to determine whether a superpixel belongs to the cliques. The second phase for detection and updating stage, through compare superpixels features surrounding region of target in training frame, with cliques, the confidence of cliques can be updated. For the target appearance model adaptive learning, a principle histogram bins clustering method be proposed to adaptive update appearance model, and the computational overhead is small. Theoretical analysis and experiments results demonstrate that our method outperforms the sate-of-the-art methods when the target under occlusion and illumination changes dramatically.
