Research On Correlation Filtering Target Tracking Algorithm Based On Non-uniform Driving

In recent years,correlation filtering technology has been widely used in the field of target tracking because of its high computational efficiency.Although most of the algorithms based on Discriminant correlation filter(DCF)have greatly improved the tracking accuracy and robustness through the construction of the model.However,as the tracking target is affected by complex background noise,the realization of high-precision and high robustness tracking in complex background is still the focus and difficulty of research.In the tracking process,due to the introduction of background information and long tracking time,the training process is also affected by boundary effect and filter degradation,which affects the robustness of tracking.In order to improve the high accuracy and robustness of target tracking in complex environment,this paper proposes two improved methods based on correlation filter tracking algorithm.The main research methods are as follows:(1)Aiming at the problems of boundary effect and response diagram distortion in target tracking based on correlation filtering algorithm,a temporal block based via time driven correlation filter(TB-TDCF)is proposed in this paper.TB-TDCF algorithm introduces the time regularization term on the basis of temporary blocks,so that the filter remains similar to the filter before multiple frames in the process of updating,improves the distortion of response diagram caused by excessive noise,reduces the over fitting defect of filter model in training,and improves the robustness of tracked filter model in the process of tracking.(2)Aiming at the model degradation problem of correlation filtering algorithm in the process of target tracking,this paper proposes a correlation filtering tracking algorithm based on relative entropy(RECF).By introducing the measurement method of relative entropy,RECF algorithm makes the correlation filter get rid of the model degradation problem caused by the simplicity of the traditional linear method in the process of model updating,and uses more complex relative entropy probability fitting to improve the robustness of the model to target tracking.In order to verify the performance of the two algorithms proposed in this paper,comparative experiments are carried out based on different data sets,and combined with 10 excellent correlation filtering algorithms in recent years to verify the excellent robustness and accuracy of this algorithm in complex background environment.