Robust Object Tracking Enhanced by Correction Dictionary

Dictionary learning method based on sparse coding has been widely used in visual tracking, since it has good performance in terms of encoding target appearance. Currently, most of the visual tracking algorithms based on dictionary learning update the dictionary with tracking results in tracking process. Consequently, the total error accumulated over time, and even cause the failure of tracking. In this paper we propose a visual tracking method enhanced by correction dictionary. We use the latest tracking results to learn the correction dictionary, which keeps the most accurate appearance information of the target at current time. In this way, our method can avoid the error accumulation. Besides, we propose a framework to compute a confidence value for correction dictionary. Experiments show the robustness of our proposed visual tracking method based on correction dictionary, especially when the target deformation occurs in the tracking process.