Correlation Filter Based Real-Time Tracking Method With Color Distribution

The visual tracking algorithm is one of the most important branches of research in computer vision.The application fields of visual tracking are becoming more and more widespread,such as pedestrian detection,intelligent surveillance,human-computer interaction,and robot.Correlation filter-based tracking is one of the most popular algorithms.This paper proposes a tracking method based on the correlation filter and solving the problems in correlation filter-based tracking.What's more,this algorithm utilizes the color distribution and the spatio-temporal context as the feature.The main work in this paper is as follows:(1)We have proposed an algorithm using color distribution and spatio-temporal context.Under the correlation filter framework,we use color distribution to alleviate the influence of similar target and the context to solve the occlusion.These two things can improve the confidence map which makes the tracking result more accurate and robust.(2)We have proposed a new update strategy and new criterion.By judging when to update,we can avoid contaminating the model and make the algorithm robust.At the same time,a scale pyramid is adopted to estimate the scale variation.(3)We have added the re-search mechanism.When the target is beyond the screen or the result is inaccurate,the tracking algorithm will detect the target.The experiment result show that the re-search mechanism improves the performance.(4)To verify the performance of the algorithm,we analyze our algorithm by comparing with other state-of-the-art methods.The experiment results are based on the CVPR13 benchmark dataset.The results indicate that our algorithm can solve many problems,including background clutter,occlusion,rotation,motion blur,deformation and so on.And the algorithm can achieve the speed at 50fps.