Abstract:  UAV (unmanned aerial vehicle) visual tracking is the core area of the future visual tracking applications, but the research is just emerging. It is difficult to track the target robustly in UAV visual tracking, due to the small and unclear appearanceof the target, the change of the flight attitude of the drone and its poor flight stability. Especially if tracking occlusion occurs, the algorithm cannot update the model after tracking drift. To alleviate these problems, we propose a multi-cue fusion tracking method with re-detection based on correlation filter. Firstly, multi-feature fusion is used to improve the discriminability of target appearance representation in UAV tracking. Secondly, when the occlusion occurs, the search area is expanded, and the sliding window sampling is used to find the target area with the highest confidence and the model update is realized. Tracking experiments on a series of challenging drone videos show that the proposed tracking algorithm has better robustness when encountering occlusion problems, which improves the accuracy of the drone in the target tracking.

Key words: UAV visual tracking, correlation filter, object tracking, multi-cue fusion