Research On Target Tracking Algorithm Using Correlation Filter And Adaptive Feature Selection

Target tracking is one of the hottest research directions in the field of computer vision,and it has a broad application prospect in video surveillance,intelligent driving,human-computer interaction,national defense and other fields.In recent years,the target tracking algorithm based on correlation filter has been paid close attention by researchers because of its lower computational cost and faster tracking speed.However,the traditional correlation filtering algorithm using single feature and weighted fusion feature will affect the whole tracking process because the feature description ability is weak and the interfered features will affect the whole tracking process.It does not perform well in some tracking scenes such as background clutter,deformation,occlusion,out-of-view and so on,therefore it is necessary to further improve the feature extracted by the target in the application scene.To solve the above problems,this paper makes an in-depth analysis in the aspects of feature selection,occlusion processing and model adaptive updating,and proposes an adaptive feature selection target tracking algorithm based on the correlation filter.The research work of this paper can be summarized as follows:First,in this paper,an adaptive feature selection algorithm based on the global confidence score in the response layer is proposed,because the lack of description ability of the correlation filtering algorithm using a single feature,and the weighted fusion feature correlation filter algorithm does not consider the problem of feature interference and destruction.The algorithm can make use of the two features in different scenes,and improves the tracking accuracy and tracking success in scenes such as background clutter and deformation.Second,a target occlusion processing algorithm based on global confidence score and local confidence score is proposed aiming at the problems of feature interference,model drift and target loss when correlation filter target tracking algorithms encounter occlusion,out-of-view and other scenes in the tracking process.The algorithm can deal with the occlusion problem according to the confidence score and update the model parameters adaptively,to improve the tracking accuracy and success in the scene of occlusion and out-of-view.In this paper,the proposed algorithm is compared with the mainstream related filter algorithms on the benchmark data containing different challenge scenarios.The experimental results show that the algorithm can well deal with the background clutter,deformation,occlusion and out-of-view scenes in the actual tracking process,and meet the real-time requirements.Compared with other mainstream algorithms,the tracking accuracy and tracking success of this algorithm show better performance.