Research On Target Tracking Algorithm Based On Kernel Correlation Filter

Target tracking technology is a research hotspot in the field of computer vision.It is widely used in military and civilian fields.As the demand for target tracking continues to increase,the problems facing target tracking technology continue to escalate,such as the interference of external factors such as changes in the size of the target object,rapid deformation,and background clutter.This all seriously affects the performance of the target tracking algorithm.Therefore,this paper put forward the corresponding improvement methods for several challenging problems of the current target tracking algorithm:1.An adaptive feature fusion algorithm for target tracking was proposed.The information of the target object can not be interpreted by a single feature,so based on the KCF algorithm to extract a single feature for target tracking,this paper proposed an adaptive fusion strategy of two complementary features,directional Histogram of Oriented Gradient and Color Names.According to the loss difference between the actual response and the ideal response,the weight is allocated and refused.The position corresponding to the maximum response value after fusion is the target position of the next frame predicted by the tracker;2.A target tracking algorithm based on occlusion re-detection was proposed.In order to solve the problem that KCF algorithm is prone to target drift or even tracking failure when the target is occluded,this paper proposed a occlusion discrimination mechanism.When the target is detected to be seriously occluded,in order to prevent the tracker from being polluted,the model is not updated at this time.When the target appears again in the video sequence,the target re detection mechanism is started;3.A scale adaptive hybrid model target tracking algorithm was proposed.The algorithm consists of two filters,one for location prediction and the other for scale estimation.The position model and scale model are updated based on the predicted target position and scale,and the optimal performance of the algorithm is used to handle target tracking in complex scenes.