Correlation Filter Visual Tracking Algorithms Based On Aberrance Repressed And Temporal Regularization

Object tracking is a popular research topic in computer vision,and it has a wide range of applications in the fields of intelligent video surveillance,intelligent transportation,autonomous driving,and human-computer interaction.For the moment,researchers have presented numerous excellent tracking algorithms,but they still face challenges when dealing with problems such as the degradation of tracking models in the time domain,the redundancy of object deep feature in the channel domain,and the interference in the object background.Therefore,to deal with the problems mentioned above,this paper is based on the correlation filter framework,and the research content is as follows:(1)Aiming at the problem that the degradation of tracking model in the time domain,aberrance repressed and temporal regularized correlation filter tracking algorithm uses the temporal regularization to constrain the object template of successive frames,which let the filter learn the similar appearance information of the object template and can adapt to the dynamic global appearance change of the object.Thereby preventing the degradation of tracking model.On OTB-2015,the proposed algorithm improves the success rate by 9.4% and the precision by 12.4%compared with the baseline algorithm.(2)To deal with the difficulty that the redundancy of object deep feature in the channel domain,channel feature selection correlation filter tracking algorithm combines hand-crafted with deep features to extract object features,not only strengthening the representation ability of the object,but also enhancing the robustness of the filter.Besides,this algorithm utilizes the channel feature selection technology to select object features,accordingly,solving the issue that the redundant feature information in the channel domain.On VOT-2016 and VOT-2018,the proposed algorithm respectively obtaines 0.439 and 0.393 scores of the expected average overlap.(3)A real-time UAV tracking algorithm based on dynamic regression is proposed to deal with the difficulty that the interference of in the object background.This algorithm dynamically adjusts the label of the tracking object in the regression equation and reduces the label value to effectively suppress the interference in the background.Compared with other excellent algorithms,the precision and the success rate performance outstanding on UAV123,DTB70,and UAVDT.And compared with other tracking algorithms based on deep network,it also shows excellent tracking performance while ensuring real-time speed.