Study And Improving On Correlation Tracking

Correlation tracking has been playing a very important role in the image-guiding tracker, it is the tracking method that search a region most approximate to the target template based on the correlation measure between two images, it avoids to segment the target or target's features from the image, and does operation only in the initial image that can retain all the information of the image, so the method can adapt to many complicated circumstances and is a feasible tracking technology.There is a systematic study on similarity measuring methods and their theories and the problem of how to update the template, including the strategy and the frequency of the update. It also analyzes the probable errors in tracking and its conundrums.To realize a real-time tracker, a multi-resolution matching method is applied to improve the matching speed, and is combined with a increasing-area search strategy to reduce the compution. The paper also introduces a new mapping function and a distance-related weighing method to reduce the influence of the noises and the non-target part in the template to get a long-time robust and precise tracking.How to solve the occlusion problem effectively has been a tough work for long. For a robust tracking algorithm, it should know when occlusion happens and how to continue to track the target correctly with the remained information all through the occlusion, and how to recapture the target after occlusion. In this paper, correlation reliability is used to judge whether occlusion happens, it also estimates the trace of the target according to the trace of former frames. Experiments show that this algorithm has a good performance.