Research And Software Implementation Of Image Matching Based On TLD Tracker

Tracking-learning-Detection (TLD) is a novel video target tracking algorithms. It selects target on the first frame, and the system will be able to initialize the tracker and detector based on the selected target, then it can track the target robustly, which has a very good research value and broad application prospect.At first, we discusses the research background and significance in this paper, reviews the research status of target algorithm domestic and aboard. Then we descript the working process of TLD framework in detail, and analyze the function of the tracking module, learning module, detection module and integration module of TLD and the related techniques used.Focuses on researching the image match of video target, and we make some improvements on traditional normalized correlation algorithm and sequential similarity detection algorithm, and propose a novel matching method, including rough matching and fine matching. We use the improved sequential similarity detection algorithm to do rough match, and then use the improved normalized cross-correlation matching method to further improve the accuracy of matching. In this paper, we realize the TLD tracking system under the environment of VS2008+OpenCV, and make contrast experiment both using the original method and the novel method. Simulation experiments show that the improved algorithm matching algorithm are more effective and superior than the TLD tracker did on the match rate, the matching speed and the successful tracking frames.