Failure Discrimination Mechanism Joint Blocking Training for Target Tracking

Recently, many traditional correlation filters have been used in target tracking with outstanding performance. However, there are still some problems such as target occlusion, target deformation and so on that need to be resolved. In this paper, firstly, the tracking failure discrimination mechanism is proposed based on the optical flow method. It compares the optical flow values of each pixel in the prediction frame with the average optical current value of entire picture. The percentage of threshold exceeded is taken as the basis for the judgment. Secondly, the filter template updating mechanism based on the idea of block training is proposed when the target tracking is judged to be a failure. When the discrimination mechanism considers that the tracking fails, the current video frame is divided and separately trained for sub-model to improve the accuracy of tracking. In order to verify the improvement effect of the algorithm, this article tested on the relatively standard datasets OTB50 and OTB100 and it can be seen in the result that our method performs better in difficult tracking environment.