Research On Target Tracking Algorithm Based On Sparse Representation And Spatio-temporal Context

Target tracking is one of the important research directions in the field of computer vision.Target tracking technology has been widely used in video surveillance,human-computer interaction,visual navigation and so on.Since the development of target tracking technology,a large number of scholars have conducted continuous research and put forward many excellent tracking algorithms.However,how to effectively solve the target appearance changes caused by occlusions,illumination change and background clutter in the target tracking process still requires further research.This paper studies on the basis of sparse representation tracking method,and improves the computational complexity and poor accuracy of target representation based on sparse representation.The main research work and innovation of the thesis are reflected as follows:For the problem of high computational complexity and low accuracy of target representation based on sparse representation tracking method,a joint inverse sparse representation and spatio-temporal context target tracking method is proposed.The merits of the inverse sparse representation tracking method and the spatio-temporal context method are mainly used.The contextual method selects candidate targets and represents the target template as a dictionary.Because candidate targets are generated by random filtering through the particle filtering method,the similarity between some candidate targets and the target template or the previous frame image tracking result is low,and the accuracy of the tracking results is affected.Therefore,the method of spatio-temporal context is used to optimize the candidate targets to improve the accuracy of the candidate target to represent target template.Finally,sparse coefficients are got according to the selected candidate target and target template through the idea of inverse sparse representation,and the optimal candidate is found as the tracking result according to the calculated sparse coefficients.Aiming at the shortcomings of the inverse sparse representation tracking method that do not make use of the similarity between the candidate targets,a target tracking method based on the group sparse model is proposed.Because there is a certain degree of similarity between the candidate targets generated by the particle filter method,in order to utilize the similarity between the candidate targets,the candidate targets are clustered using the K-means clustering method according to the similarity.The candidate target with the group structure obtained is used as a dictionary to represent the target template.Finally,by solving the group sparse coefficients and selecting the most similar one or more groups of candidate targets from the candidate targets of the group structure,the most similar candidate targets are selected as the tracking results.