Object Tracking Algorithm Based On Bidirectional Sparse Representation

Visual target tracking as a cutting-edge technology in the field of machine vision,it is an important part of the component of artificial intelligence,mainly related to image processing,signal processing,pattern recognition and other professional knowledge,and it has been used widely in security monitoring now,visual inspection and smart driving and other emerging industries.so visual target tracking technology will have great research and practical value in the future.In this paper,on the premise of solving the sparsity of one-way sparse representation model,because the one-way sparse coefficient matric is solved separately,they can not excavate effectively the correlation between the positive and reverse sparse coefficient matrices.So the sparse representation based on bidirectional target tracking algorithm is proposed,and the bidirectional sparse object representation is introduced into the Bayesian tracking framework,so that it can better reduce the adverse effects of changes in the appearance of the target.The main research work of this paper is as follows:(1)Research on related basic theories of target tracking at home and abroad,we also explore and implement sparse coefficient solving algorithm based sparse representation at the same time,so it provides some basic theoretical knowledge for the research of the target tracking algorithm proposed in this paper.(2)One-way sparse representation fails to correlate their sparse coefficients,in order to better express candidate samples in the template set,a bidirectional sparse representation target tracking algorithm framework is proposed,L2 norm is used to constrain the positive and negative reconfiguration errors,The positive and inverse sparse coefficient matrix is optimized by accelerating approximation gradient algorithm.through the weighting matrix combined with positive and inverse sparse coefficient matrix,the candidate samples with the maximum difference degree on the candidate sample set are used as the tracking optimal target,the proposed algorithm is much more stable than the traditional one-way sparse representation tracking algorithm.(3)In the similarity distance measure between target template set and candidate template set,because the weight of traditional Euclidean distance is inaccurate under the influence of target occlusion,illumination and shadow,therefore,an improved local weighting distance metric algorithm is proposed.The improved metric algorithm runs on the video sequence of complex environment and has a high robustness compared with the traditional Euclidean distance metric algorithm.