Abstract: Taking the gradient features and color features of an image into consideration, an improved visual tracking algorithm based on correlation filter tracking is proposed in this paper. The algorithm uses the Bayesian theory to model the color information, and carries out object tracking with the combination of gradient features’ correlation filter output and object confidence integral map which is obtained from dense object posterior probability. Meanwhile, the algorithm conducts a quality assessment of the results of the object tracking. Once the tracking quality turns out to be unreliable, the object re-detection process will start, which is based on the object confidence integral to determine the candidate objects. As to the video frame with unreliable tracking quality or no reliable object after re-detection, the tracking model will not be updated online. Experiments show that the proposed algorithm can effectively avoid the problem of unreliable tracking and model drift caused by circumstances varying, and its tracking performance is obviously improved compared with those of several state-of-the-arts correlation filters.

Key words: correlation filter, visual tracking, object re-detection, object confidence integral