Research On Target Tracking Algorithm Based On Kernel Correlation Filter

As an important part of computer vision,target tracking is widely used in autonomous driving,surveillance security,human-computer interaction and military fields.In the actual tracking process,the target often undergoes internal changes such as deformation,scale change,etc.,and is often affected by external influences such as illumination,occlusion,and background clutter.Therefore,it is important to enhance the robustness of the target tracking algorithm in various scenarios.In this paper,based on the kernel correlation filtering algorithm,an in-depth study is carried out,and some effective improvement methods are proposed for the difficult problems in current target tracking.The specific research work is summarized as follows:1.The mathematical principle of kernel correlation filtering algorithm is introduced,and the parts of tracking algorithm based on kernel correlation filtering are introduced in detail,and the shortcomings of current algorithm are analyzed.2.In order to enhance the robustness of the target model and solve the target scale change in the target tracking process,a scale-adaptive kernel correlation filtering target tracking algorithm based on feature fusion is proposed.The algorithm uses the kernel correlation filtering algorithm as the basic framework,and uses the OPP color feature and the HOG feature to fuse the fusion feature in the feature extraction stage.The fusion feature is used to enhance the representation of the target appearance and improve the robustness of the tracking algorithm.In addition,the different scales of the target are calculated and the scale with the largest response value is selected as the target scale to realize the adaptation of the target scale in the tracking process.The experimental results show that the proposed algorithm can better adapt to the scale change of the target and can obtain good tracking effect in various challenge scenarios.3.In order to cope with the influence of the external environment of the tracking process and the changes of the target itself,a target tracking algorithm of the joint model is proposed.The target tracking is performed by using the key point tracking algorithm and the kernel correlation filtering algorithm.When the target is occluded,the key points of the target are filtered by the forward and backward error,and the influence of the background can be reduced by using the key point information.In addition,the Gaussian mixture model is used to process the samples to generate different components.Each component corresponds to a group of relatively similar samples,which reduces the sample size while maintaining the sample diversity and preventing the algorithm from over-fitting.The experimental results show that the proposed algorithm still has a good tracking effect in complex environments.