Research On Multi-strategy Fusion Moving Target Tracking Algorithm Based On Correlation Filtering

Tracking of moving targets has always been a hot topic widely studied and discussed by scholars.In recent years,the moving target tracking method based on correlation filtering has shown excellent tracking performance,but still faces several challenging problems,among which the more prominent problems are reflected in illumination changes,target deformation,occlusion and complex scenes.These problems can lead to tracking failures,making it difficult to achieve robustness and accuracy in moving target tracking.Aiming at the problems encountered by the moving target tracking algorithm in complex scenes,this thesis improves on the basis of correlation filtering to form a multi-strategy fusion moving target tracking algorithm.The algorithm of this thesis mainly improves and optimizes CSK(Circulant Structure Kernel,CSK)algorithm and CN(Color Name,CN)algorithm from two aspects.The specific work is carried out in the following aspects:(1)Firstly,the status and significance of the correlation filtering algorithm are introduced.Then the tracking process of the correlation filtering is elaborated,including feature extraction,training filter,target detection and filter update.Explain in detail the problems that exist in them.(2)Problems with respect to algorithm performance and target scale changes.Since the original CSK algorithm and CN algorithm can't judge the sequence color space and target scale well,which leads to the degradation of tracking performance,this thesis uses a strategy of integrating two original algorithms,making it a more complete algorithm.When facing video sequences of different color spaces,it is possible to adaptively select the optimal tracking method for tracking;for the above two algorithms can not adapt to the problem of target scale change,this thesis introduces DSST(Discriminative Scale Space Tracker,DSST)algorithm The strategy of mesoscale filter is to design a target tracking algorithm with color space and scale change to ensure that the proposed algorithm can adapt to the change of target scale,making tracking more robust.(3)Partial occlusion of this problem for the target.Since the target tracking algorithm of color space and scale change does not solve the mechanism of target occlusion,the target fails when it is partially occluded.In this regard,this thesis introduces the Peak to Sidelobe Ratio(PSR)value strategy to control the update of the target template,and the learning rate of the target template update adopts an adaptive strategy to design a lead-in to the peak.Correlation and correlation tracking algorithm for valve ratio and learning rate adaptive,and qualitative and quantitative analysis using standard test set.Experiments show that the improved algorithm can accurately track the target in the complex background such as target scale change and occlusion.The tracking performance is better than other comparison algorithms and has better robustness.