This paper studies the concept drift detection algorithm,and modify the traditional concept drift detection algorithm to make it still effective in the visual tracking scenario.Our tracking algorithm combines sparse representation with concept drift detection to form a robust tracker.Finally,our experiments show that our tracking algorithm achieves the state-of-the-art accuracy.In the field of data mining,the old model may become invalid due to the change of context.So the purpose of concept drift detection algorithm is to detect these changes and update old model accordingly.However,the usual concept drift detection algorithm lack the ability of handling visual tracking scenario,because of its complexity.To overcome this,we introduce the Wilson interval score and propose a novel concept drift detection algorithm that is effective in visual tracking.The proposed tracker in this paper is discriminative.A big issue with discriminative tracker is that it usually degenerates along with the tracking process because of the accumulated errors.Our concept drift detection algorithm can fix this issue well.And furthermore,we introduce the sparse representation to improve the robustness of our drift detection algorithm.Finally,we show the accuracy of our tracker by massive experiments and conclude our current work along with future works. |