Research On Video Target Tracking Algorithm Based On Sparse Representation

Video target tracking is a method to estimate the target position in each frame of video.It is widely used in traffic and other fields.Researchers at home and abroad have done a lot of research on video target tracking,and put forward a lot of effective tracking algorithms.However,video target tracking becomes more complicated because of the heavy occlusion,sharp Illumination variation,and target deformation and so on.In order to improve the accuracy and robustness of the video target tracking algorithm,this paper has carried out a thorough research on this problem,and the specific contributions are as follows:(1)A multi-feature visual tracking algorithm based on weighted local cosine similarity is proposed.The algorithm uses RGB color histogram feature and gray sparse representation coefficient feature to build the appearance model,which effectively enhances the color description ability of the appearance model.In addition,the weighted local cosine similarity measure function is used to fuse the two features of RGB color histogram and gray sparse representation coefficient.A multi feature fusion model is built,and the corresponding template update strategy is designed.The accuracy and robustness of target tracking is greatly improved.(2)A target tracking via quaternion-based sparse representation for color images is proposed.Based on the quaternion theory,the algorithm is designed for the sparse representation framework of color images and the corresponding appearance model is built.Meanwhile,in order to improve the accuracy of tracking,based on histogram crossover function,an observation model suitable for quaternion-based sparse representation is designed,and the target model is updated initiatively to adapt to the appearance change of tracking target.In summary,based on sparse representation theory,two video target tracking algorithms have been implemented,and experimental analysis has been carried out.The experimental results show that the two tracking algorithms proposed in this paper achieve accurate and robust tracking effect and reduce tracking drift.