Object Tracking Algorithm Based On Siamese Network With Drift Correction Mechanism

As a basic research problem in the field of computer vision,object tracking has aroused extensive research interest of a large number of researchers,and a large number of excellent algorithms have been proposed one after another.At present,object tracking algorithm has been widely used in video security,intelligent transportation,automatic driving and other fields.In the actual application scenario,the target tracking process is often accompanied by problems such as illumination change and background interference,which will lead to serious drift of the tracker when tracking the target.The drift problem refers to the deviation of the positioning frame from the target.How to effectively solve the drift problem in tracking is one of the difficulties in object tracking research.Therefore,under the framework of deep learning,this paper makes a detailed study on the drift problem in the tracking process:Firstly,aiming at the problem of insufficient tracking accuracy of the tracker itself,a Siamese network target tracking algorithm integrating multiple attention is proposed.This algorithm modifies the backbone network on the tracker based on Siamese network and integrates the "multiple attention algorithm" proposed in this paper to extract features with stronger semantic expression ability,so as to improve the accuracy of tracking.Among them,multiple attention algorithm combines two visual attention mechanisms: soft attention and self attention.It can not only highlight important features,but also capture long-range dependent information.Secondly,various current tracking algorithms are deeply studied.Aiming at the current situation that the tracker can not correct the existing drift problem,a drift correction algorithm based on target matching and detection is proposed.It can make the target tracker have the ability to monitor the drift problem in the tracking process,and actively correct the problem after finding the drift problem.That is,it has the ability to relocate the target position,so as to improve the performance of the tracking algorithm.Finally,the two methods are combined to achieve a better object tracking algorithm,and detailed experimental verification and performance analysis are carried out under the pytorch deep learning framework.