UAV object tracking via the correlation filter with the response divergence constraint

Aiming at the problem that targets are easily subject to deformation and background clutter interference in the drone sequences, this paper proposes a novel unmanned aerial vehicle (UAV) object tracking method based on the correlation filter with the response divergence constraint. According to the consistency of the filter variation between the previous frame and current frame, the response divergence of different filters acting on the same training sample is modeled. Furthermore, an objective function with the constraint mechanism is built, which can learn the target variation accurately and promote the robustness of filters. Meanwhile, an auxiliary is introduced to construct the optimization function. The alternating direction method of multipliers is used to optimize the solution of the filter and auxiliary variable. We have tested the proposed algorithm and eleven state-of-the-art algorithms on three UAV video databases including DTB70, UAV123@10fps and UAVDT. Experimental results demonstrate that our method is superior to comparison algorithms on two evaluation indicators such as tracking accuracy and success rate and has good robustness for illumination variation, deformation, occlusion, motion blur and other challenging attributes in complex environments from the view of UAV. Meanwhile, the average tracking rate of our algorithm reaches 21. 7 frames per second, which meets the real-time requirements of UAV. © 2021, The Editorial Board of Journal of Xidian University. All right reserved.