A robust tracking algorithm based on sparse histogram

In order to strengthen the robustness of video object tracking and overcome the drastic appearance changes caused by diverse challenges, a generative algorithm based on sparse histogram was proposed. Firstly, the image was partitioned into patches by using overlapped sliding window. Secondly, the local features of the object were extracted, and the sparse representation was exploited to achieve the object coefficient vector. Here the Augmented Lagrange Multiplier (ALM) was introduced to solve the h norm minimization problem, and then the occlusion information of the object was computed to construct sparse feature histogram to be the templates. Finally, the candidate state which had highest similarity score computed with reference templates was chosen as the best result within the particle filter framework. In addition, a simple yet efficient update mechanism was formulated to update occlusion information to construct new template histogram, which handled the variation of the object appearance. The tracking results on diverse testing video sequences show that our algorithm has a better peformance than other approaches.