Combining color feature for scale adaptive correlation filter tracker

摘要

In order to improve the robustness of correlation filtering(CF) tracker, and solve the problems that the traditional CF methods cannot overcome the change of the object scale and the color information was not used. A scale adaptive correlation filtering tracker based on color feature fusion is proposed. Firstly, we transform the target search area of the image from RGB color space to Lab color space, and then extract the three channel Lab color features of the search area. Secondly, we combine the color feature and HOG feature into multi-channel feature. The multi-channel correlation filter was employed to locate the center of the target. Finally, the target scale model is established by Lab color feature. Acquiring different scales image block from the target position at the current frame. We get the optimal target scale by comparing with the scale model. Experiments were performed on 35 public benchmark challenging color sequences and compared with other 5 tracking methods. The results show that our proposed approach outperforms state-of-the-art tracking methods, and has a real-time tracking speed of 66 frames/s.