Research On Visual Tracking With Siamese

Object tracking is a task in computer vision that has received a lot of attention from researchers.Object tracking have applications in several fields.In video sequences,the task of Object tracking is to obtain information about the Object in the initial frame and to predict the position of the Object in subsequent frames.Since current trackers are not very accurate and operate slowly,the study of Object tracking techniques is still important.The following work has been carried out to address the problems of a single feature extraction network,simple or redundant template updates and confusing information fusion in the tracker.1)A Triplet feature extraction network is proposed to address the problem that the feature extraction network in the tracker only use the initial template and the search region.A supplementary template branch is added to the Siamese network to form a Triplet network.Through the analysis of experimental results,the problem of insufficient template information is effectively solved.2)A new template fusion method is proposed to address the problem of missing or confusing template information in the template update method of the tracker.Supplementary templates are embedded into the tracker using the template fusion method.The improvement was carried out on several trackers and the results showed that the improved tracker improved the tracking accuracy of the original tracker by 2.2%.3)The information fusion method of graph attention is used to address the problem of the large amount of mutual correlation calculation and the introduction of large amounts of background information in information fusion.The obtained features are fused using the graph attention module.The evaluation was carried out on several datasets and the results showed that the proposed tracker improved the accuracy by 1.9%over the baseline tracker.The use of Triplet network instead of Siamese network,the use of fusion to embed supplementary templates into the tracker,and the use of graph attention module for information fusion enhance the performance of the tracker and provide a new research direction for the development of the Object tracking.Figure 18;Table 9;Reference 59...