Research On Visual Tracking Via Low Rank Projection And Sparse Representation

In recent years,with the development and popularization of computers and related hardware equipments,single target tracking technology is widely concerned by increasing researchers.Single target tracking is an important part of computer vision in the current state,its robustness is always restricted to target occlusion,illumination change,target pose change and so fourth,then the application is also limited.On the basis of previous work of the visual tracking,this paper researches on rank minimization theory and sparse representation theory,then incorporate these theory into the target tracking field,and major contributions are showed as follows:1.The current statues and relevant theories of visual tracking.Firstly,the research status of visual tracking is analyzed and discussed;and then this paper introduced the low rank matrix recovery theory and sparse representation theory in detail,then derived the solving method of each model;lastly,many classical algorithms and relevant evaluation index are described and analyzed.2.Robust visual tracking based on analysis of sparse error matrix in low rank projection.After studying the target tracking technology based on sparse representation theory,a visual tracking algorithm based on analysis of sparse error matrix in low rank projection is proposed.During processing,the original data is decomposed into low rank component and sparse component,and then sparse error matrix is obtained by candidate target projecting to mapping space based on relationship between the original data and the low rank component,at last,best candidate target could be chosen by making using of sparse error matrix.In order to overcome the effect of model drifting,target templates are updated dynamically with the similarity of target templates and candidate targets,when the similarity is high,the algorithm updates target template by using the candidate target,otherwise,low rank component of candidate target is used to update the target template.On the basis of above theory,the algorithm is achieved by using overall prediction framework based on partial filter.The experimental results on several sequences show that this algorithm has better performance than that of the state-of-the-art tracker.3.Robust visual tracking via incremental subspace learning and local sparse representation.The algorithm based on incremental subspace model is sensitive to target occlusion,so this paper proposed an improved visual tracking algorithm based on incremental subspace learning and local sparse representation.The algorithm adopts local sparse representation to test occlusion and rectifies the incremental subspace learning according to the occlusion detection outcome.During the decision process,the candidate target is chosen by the collaborative model of incremental subspace learning and local sparse representation.In the frame of partial filter,target tracking is achieved by different update strategies.The experiments are carried out on several challenging image sequences and compared with the existing tracking algorithms,results demote that this algorithm has good robustness.