Tracking The Video Compression Algorithm Based On Perception

In the field of computer vision and image processing, visual object trackinghas been developed extremely in military affairs and civil Engineering. Manyvisual object tracking algorithms have been proposed by researcher in the fieldof object tracking, however there are lots of factors will be encountered in theactual tracking process, such as object occlusion, appearance change,illumination variation, target and background have similar patterns etc. So,object tracking is a challenging task to develop effective and robust algorithms.The target tracking based on sparse representation method is an importantpart of this thesis, with the objective of improving accuracy and robustness ofvisual object tracking algorithms, we also study appearance model based onsparse representation, the theory of Compressive Sensing (CS) and constructclassifier. As the main challenge for visual object tracking is to account forobject appearance change and cluttered background，in this paper we propose arobust appearance model which exploits both holistic and local templates, inorder to cope with situation of object occlusion and cluttered background. Thedimension of the sample image is reduced via compressive sensing account forreduce computational complexity. In the discriminative classifier module, weintroduce kernel function method to improve tracking performance in thesituation of target and backgrounds have similar patterns, this method guaranteestracker has long-term stability and better accuracy. This algorithm based onBayesian inference and kernel sparse representation complete object trackingtask. Both qualitative and quantitative evaluations on standard video testsequences demonstrate that the proposed algorithm performs favorably.