Research On Tracking Algorithm Based On Local Sparse Representation Template Matching

Tracking technology is an important technology in the field of security,many scholars are studying the tracking technologies and designed tracking systems,such as intelligent monitoring system,burglar alarm system,video retrieval system and intelligent transportation system,and several types of algorithms were also proposed.This paper mainly studied algorithms based on sparse representation model,particle filter and sparse representation(PFSRA)tracking algorithm and local sparse representation(LSRTA)tracking algorithm were analyzed through simulation experiments.It is found that the tracking effect of LSRTA algorithm was better.But the target will be offset in the case of large area occlusion,fast moving target and big change of the target appearance.Based on the LSRTA tracking algorithm,a local sparse representation template matching(LSRTMTA)tracking algorithm with correction mechanism was proposed.Experimental results show that the proposed algorithm not only retains the advantages of LSRTA tracking algorithm,but also has the automatic correction deviation function,improved tracking effect and increased fault tolerance in the target tracking process.The main work of this paper were as follows:1.Several algorithms for sparse representation were summarized.By comparing the BP algorithm,OMP algorithm and BOMP algorithm to analysis the advantages and disadvantages of the three algorithms.PFSRA tracking algorithm and LSRTA tracking algorithm are introduced.experimental results show that LSRTA tracking algorithm is faster and more accurate than PFSRA tracking algorithm.However,when the target is in large area occlusion and fast moving,using LSRTA tracking algorithm to track,the target offset will occur.2.Aiming at the target offset problem,LSRTMTA tracking algorithm with modified mechanism was proposed.In the LSRTA algorithm tracking process,the algorithm determines whether the offset occurs by continuously calculating the matching value of the new template with the current template.When the target is offset,the tracking of LSRTA algorithm is stopped.Target is repositioned by matching between template and frame image.When the target position is determined,the tracking of LSRTA algorithm is carried out again.3.An algorithm running platform realized.The platform can read the network camera video information and intuitive display of the tracking effect of the tracking results about PFSRA tracking algorithm,LSRTA tracking algorithm and LSRTMTA tracking algorithm.