Research Of Object Tracking Based On Correlation Filter

Object tracking is one of the most challenging problems in computer vision,researchers have proposed a variety of excellent object tracking algorithms in recent decades.However,it still face with a series of challenges in the real environment.The appearance of object changes,including scale changes,illumination,rotation,occlusion and so on,will lead to the object failed to track.In order to solve these problems,object tracking based on correlation filter has been researched and explored in this paper.The contributions of this paper are as follows:First,a scale and rotation adaptive tracking method based on kernelized correlation filter is proposed.In order to solve the problem of scale changes and rotate during tracking,first,the algorithm determines the center position of the object via kernelized correlation filter.Then the algorithm estimates the scale changes and rotation angle of an object using keypoints matching.In the process of keypoints matching,the method eliminates unstable keypoints using forward and backward matching.The next,the algorithm estimates the ideal scale and angle by considering the weight of keypoints.At last,the method detects whether the target is occluded,and then update keypoints set and object model more reasonable,and hence improving the robustness of the algorithm.Second,an adaptive weighted fusion method based on color names feature and HOG feature with multiple kernels and multiple channels based on correlation filter is proposed.The algorithm first train kernel ridge regression classifier using two kinds of features respectively,instead of only one feature in kernelized correlation filter.Then,according to the magnitude of the response values to determine the center position of the object,the weights of the complementary kernel features and updating model are adaptively assigned,improving the robustness of kernelized correlation filter.Finally,the above algorithms together to form the final algorithm.In the experiment,the algorithm use ten standard videos sequences with differentinterference factors to test,including occlusion,scale changes,rotate.The results of experiments show that the proposed algorithm not only can adapt to changes in the target appearance under complex scenes,but also completely meet the tracking demand of real-time scenario.