Optimization Of Object Tracking Algorithm Based On Video Sequences

Object tracking of video sequences is one of the key technologies in computer vision, and is the basis of advanced machine vision, which has extensive applications. However, with the interferences of the target appearance change, complicated environment and the change of the target scale, object tracking algorithms in practical application still exist some problems such as drifting, false tracking and so on. To deal with the above mentioned problems and improve the efficiency of the algorithm, the mean shift tracking algorithm and the compressive tracking algorithm based on the region feature are studied and optimized in this paper. The main work is as follows:Based on the mean shift algorithm, the paper firstly proposed a double weights algorithm in order to solve the problem of interference of the similar color background. Through combining the background weaken weight with the object center weighting when modeling the target, the anti-interference performance of the algorithm is improved. Secondly, to improve the tracking robustness on the influence of occlusion, illumination, deformation, a template update strategy is proposed. Finally, in order to solve the problem of localization error and even tracking failure in long time tracking caused by the fixed tracking window size, a method adopted by the CAMShift algorithm is used to update the tracking window which through zero and first order moment of back projection map.Based on the compressive sensing theory, a sample selection method and a learning factor updating strategy are proposed to settle the influence of occlusion, appearance variation and object movement speed on the compressive tracking algorithm. The optimized algorithm, based on the idea of kernel function, assigns the different positive samples with different weights according to their distance to the target center. Then the difference of positive samples is enhanced. At the same time, the Bhattacharyya coefficient relationship between before and after two frame images is adopted to auto-update the learning factor of model parameters.The experiments demonstrate that the optimized algorithms have a good tracking real-time performance and enhance the effectiveness of tracking result as well.