Matting and super-pixel based target tracking algorithm

Model updating is a critical problem in target tracking. Inaccurate foreground and background estimating will degrade the tracking performance even cause drift problem. In order to address this problem, a robust tracking algorithm based on super-pixels and Matting is proposed. We use feature matching and color-histogram of super-pixels to offer enough foreground and background information for Matting. In addition, we sample the patches of object to record the appearance information which can deal with the situation of occlusion. Compared with other tracking methods, experiments show that our algorithm can overcome the problem of model drift and track the object with better accuracy.