AUTHOR=Cui Jianjie , Wu Jingwei , Zhao Liangyu 

TITLE=Learning channel-selective and aberrance repressed correlation filter with memory model for unmanned aerial vehicle object tracking

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 16 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2022.1080521

DOI=10.3389/fnins.2022.1080521

ISSN=1662-453X

ABSTRACT=In order to realize that computers can accomplish specific tasks intelligently and autonomously, it is common to introduce more knowledge into intelligence (AI) technology as a priori information by imitating the architecture and mindset of human brain.Currently unmanned aerial vehicle (UAV) tracking plays an important role in military and civilian fields.However,robust and accurate UAV tracking remains a demanding task due to limited computing capability,unanticipated object appearance variations and volatile environment.Here,inspired by memory mechanism and cognitive process in human brain and considering the computing resources of the  platform,we propose a novel Discriminative Correlation Filter (DCF) based trackers tracking method with memory model by introducing dynamic feature-channel  weight and aberrance repressed regularization into the loss function and adding an additional historical model retrieval module.Specifically,the feature-channel  weight integrated into the spatial regularization gives the filter the ability to select features.The aberrance repressed regularization provides potential interference information to the tracker and is advantageous in suppressing aberrances  caused by both background clutter and appearance changes of the target.Through the joint optimization of the above two,the proposed tracker could not only suppress the potential distractors,but also train a robust filter by focusing on more reliable features.Furthermore,the overall loss function could be optimized using Alternative Direction Method of Multiplies (ADMM) technique method,which further improve the calculation efficiency of the algorithm.Meanwhile,the historical model retrieval module encourages the tracker to adopt some historical models of  past frames in video to update the tracker and make full use of historical information to construct more reliable target appearance representation.We evaluate our method on two   challenging UAV benchmarks.The results have demonstrated that our tracker achieves superior performance against other state-of-the-art tracking algorithms.