Research On Stable Target Tracking Technology Based On Correlation Filtering

In recent decades,as a hot research direction in the field of computer vision,target tracking has made breakthrough progress,and it has also been applied in military guidance,human-computer interaction,and intelligent monitoring.However,because there are too many challenges to be solved by target tracking,how to track targets stably under various challenges remains a problem.In recent years,the correlation filtering based tracking algorithm has attracted the attention of a large number of researchers with its good tracking accuracy and amazing tracking speed,making the development of related filtering tracking algorithms change with each passing day.But the related filtering algorithms still have a lot to study.This paper focuses on the shortcomings of the kernel-related filter tracking algorithm,and proposes corresponding solutions and experiments.The work of this paper is summarized as follows:Firstly,in order to solve the scale estimation problem of the kernel correlation filter tracking algorithm in tracking,a scale-adaptive kernel correlation filter tracking algorithm is proposed,which decomposes the target tracking task into two large blocks: the prediction target position solution and For the scale solution,two large blocks are learned by the regularized least squares classifier to obtain two correlation filters.The position filter obtains the target position information,the scale filter obtains the target scale information,and then updates the two filters according to the tracking result..The experimental results show that the algorithm can solve the problem of target scale change better.Secondly,in order to track the instability of the kernel correlation filter tracking algorithm under occlusion,an anti-occlusion kernel correlation filter tracking algorithm is proposed.The algorithm mainly determines whether the target is occluded by setting the threshold value of the peak sidelobe ratio and the correlation peak average energy of the response graph.If it is occluded,the template parameters are not updated.This can suppress the influence of occlusion on the model and achieve stable tracking of the tracking algorithm under occlusion.Experiments show that the algorithm has certain resistance to occlusion.Thirdly,in order to solve the problem that the single HOG feature has weak description ability to the target and the tracking ability of the tracker in the complex environment is weak,a kernel correlation filter tracking algorithm based on feature fusion is proposed,and a Bayesian classification based on the classification is proposed.The color statistics feature.The algorithm first uses the two-feature-based tracker to perform target tracking separately,and then combines the two tracking results in the decision-making layer,so that the tracking algorithm can adapt to the complex conditions such as target deformation,illumination change,motion blur and so on.Experiments show that the algorithm can stably track many complex situations.