Abstract: Based on multiple appearance models， a novel multiscale mean shift tracking algorithm was proposed to deal with the problems caused from the relative simplicity of the single appearance model and the absence of the update strategy under the original framework of mean shift tracking. Based on the multiple appearance models which can be obtained by using sparse principal component analysis， numerous converging points were located by running the basic mean shift trackers in parallel. The weighted center was calculated by setting the converging points as the candidate particles， and the best particle was chosen to determine the current state of the object. Experimental results showed that the proposed method was more robust and stable against pose variation， background clutter and occlusion in comparing with other competing tracking models.

Key words: object tracking, mean shift, sparse principal component analysis, adaptive update, weighted center