Research on adaptive target tracking algorithm

Recently, many tracking algorithms based on correlation filtering are proposed, and good performance is obtained.. Although these algorithms have extremely high accuracy, they cannot solve the tracking failure caused by cumulative errors and scene changes. So how to design a robust tracker is very import. An adaptive target tracking algorithm is proposed to achieve fast and robust tracking. For feature selection, we propose a hybrid feature based on HoG, CN, and Haar-like to express the tracked target from multiple angles such as color, edge, and detail. In order to solve the speed reduction caused by complex fusion features, we preset a simple classifier in the front end of the algorithm. The role of the classifier is to distinguish the difference between the target and the background, and to determine whether to use fusion features for tracking. Finally, we use simple mobile prediction to reduce the impact of accumulated errors. To evaluate the proposed algorithm, a wide range of experiments are performed using some of the OTB benchmarks.