Target Re-detection And Tracking Based On Sparse Representation And Correlation Filtering

Video target tracking technology has always been an integral part of computer vision research.As the main aspect of artificial intelligence,computer vision has played a key role in promoting its development.Vision target tracking technology is widely used in the fields of defense and military,biomedicine,drone reconnaissance,and intelligent automatic driving.At present,the visual target tracking technology faces some difficult problems.When the target moves rapidly,the surrounding scene changes complicatedly,the target deforms and the target is blocked seriously,the tracking frame will be different from the real position of the target,or even lose the target.Therefore,this paper proposed the Target re-detection and tracking based on sparse representation and correlation filtering to deal with the serious occlusion problem in the process of target tracking.This algorithm improved the robustness and accuracy of visual target tracking,and can track in real time.The specific work of this paper was as follows:.First,this paper briefly introduced the long-term tracking algorithm(LCT)and the target tracking algorithm based on sparse representation which explained the operation process of the LCT algorithm in detail,studied the application of two independent correlation filters,translation and scale,and learned about the sparse representation model and the application of sparse representation in target tracking.Then the algorithm operation process and the principle of particle filter motion model were analyzed.The advantages and disadvantages of the LCT and the target tracking algorithm based on sparse representation model were briefly analyzed.Secondly,aiming at the problem of target tracking failure caused by serious occlusion,this paper combined two correlation filters and particle filter tracking algorithm based on sparse coding appearance model.And the algorithm in this paper introduced the average peak correlation energy value(APCE)evaluation target response confidence map to determine whether the target needs to be relocated.The particle filter was used as the motion model for the second estimation of the target position,which increased the accuracy of the algorithm.The sparse representation was used as the appearance model to process the occlusion part,which better solved the problem of serious occlusion of the target.Finally,the algorithm was tested on more than ninety video sequences of object tracking benchmark.Comparing with some related filtering target tracking algorithms through qualitative and quantitative analysis,the algorithm in this paper had better accuracy and robustness on the premise of real-time requirements.And this algorithm can better solve the problem of tracking failure when the target was seriously occluded.