Adaptive Model Updating Correlation Filter Tracker with Feature Fusion

Abstract

Aiming at the poor accuracy of a single feature in the challenging scenarios, as well as the failure of tracking caused by partial or complete occlusion and background clutter, a correlation filter tracking algorithm based on feature fusion and model adaptive updating is proposed. On the basis of the background-aware correlation filter, the proposed algorithm firstly introduces the CN feature and integrates with the HOG feature to improve the accuracy of tracking. Then, the Average Peak-to-Correlation Energy (APCE) is introduced, and the results of object tracking are fed back to the tracker through the ratio changes. The tracker is adaptively updated, which improves the robustness of the algorithm to occlusion and background clutter. Finally, the proposed algorithm is experimented on the self-build ship dataset. The experimental results show that the algorithm can adapt well to complex scenes, such as object occlusion and background clutter. Compared to the state-of-the-art trackers, the average precision of the proposed tracker is improved by 2.3%, the average success rate is improved by 2.9%, and the average speed is about 18 frames per second.