Research On Multi-target Tracking Algorithm Base On TLD

Video target tracking plays an important role in image processing and is widely used in intelligent monitoring system. In order to solve the problems of the object disappeared and obtaining accurate tracking target. An on-line learning TLD algorithm was proposed by Z.Kala in Czech. This paper strudies this TLD algorithm deeply and mainly introduces the framework and designing of the algorithm especially for the detection,tracking, learning and integral modules.On the basis of analyzing and researching this TLD algorithm,we found that this algorithm can achieved good performance in the deformation and light interference. But the traditional TLD algorithm can get good performance in single target and is invalid for multiple targets. To sovle this problem,a improved algorithm of tracking target based on TLD,which is valid for multiple targets,is presented. This paper gives a framework of the multiple targets tracking in practical application. We also present the detailed design for the multiple targets tracking.The main contents are as following: Using the new data structure to store the data and using the new ways to select the feature points in mult-target tracking. In order to reduce the number of multi-objective sliding window in the detection module, we normalized the size of the multiple target frame. We re-design the implement of variance classifier and ensemble classifier in the detection and learning modules. Finally,this paper describes the accuracy and effectiveness of this algorithm and gives the advantages and disadvantages. It indicates the direction for the future study.