Research On Correlation Filtering Algorithm Based On Memory Mechanism For Target Tracking

With the development of computer technology,the problem of target tracking has increasingly become a hot issue in the field of computer vision,and is the basis for other subsequent processing operations.In most target tracking scenarios,a variety of conditions can make many tracking algorithms extremely challenging.At present,the correlation filtering algorithm in the tracking algorithm is very fast,but the accuracy is not high.The deep learning algorithm has high accuracy,but the real-time performance is not up to the requirement.Therefore,it is still very important to study target tracking algorithms.Aiming at the challenges of background confusion,rapid target movement,deformation and so on,the subject is devoted to the research of memory mechanism-related filtering algorithms for target tracking.The main work is as follows:1)In the target tracking process,contextual information is added to the relevant filter,and color features and directional gradient features are extracted to construct the target appearance model.In the target model update process,the peak sidelobe is used to update the tracking parameters.The algorithm is tested on OTB50,OTB100,UAV123,TC128 datasets,and it has a high success rate and accuracy.2)In order to make up for the inaccurate tracking problem caused by a single feature,the CN feature,HOG feature,and texture feature of the target are extracted after the target is initialized to construct the appearance model.A three-layer rotating circle memory model is used to update the target template.The algorithm is tested on OTB50,OTB100,UAV123,TC128 datasets,and it has relatively strong tracking ability.3)In order to prevent a sudden change in the response graph during the relevant filtering operation,a filter that can suppress anomalies is constructed,and the ADMM method is used to speed up the calculation of the filter.The CN feature and HOG feature are extracted to construct the appearance model.And a scale filter is added to the method.During the update process,the memory model is used to update the target model and the scale filter is set to reasonable parameters.The algorithm is tested on OTB50,OTB100,UAV123,TC128 datasets,and it can achieve more accurate tracking.The method proposed in the article provides a direction in the field of target tracking.Figure 29;Table 11;Reference 58...