Research On UAV Target Tracking Based On Dynamic Feature Weight And Spatial Regularization

Visual object tracking is one of the important branches in the field of computer vision.The task of object tracking is to continuously estimate the center position and scale information of the target in subsequent image sequences based only on the given target information in the first frame.In recent years,the performance of unmanned aerial vehicles(UAVs)has been rapidly improved with the development of hardware,and its excellent adaptability has made it widely used in aerial photography,traffic patrols,automatic tracking,and pedestrian detection.Existing object tracking algorithms have achieved good tracking accuracy and success rate in static scenes.However,in real scenes,complex factors such as occlusion,background disturbance,similar targets,and lighting changes make it still a great challenge to design a robust tracking algorithm.Based on the existing correlation filters,this thesis focuses on how to carry out effective feature selection and how to enhance the spatial regularization of the model.The specific work and innovation of this thesis are as follows:(1)A robust UAV vision tracking method with dynamic feature weight selection is proposed.The feature constraints and representation ability are enhanced through dynamic feature weight selection.Firstly,a feature weight pool is defined to store different combinations of weights for multiple features.Then,the feature pool item is introduced into the correlation filter model,and the model is optimized using the alternating direction multiplier method.Additionally,the fast Fourier transform technique is used to convert the spatial domain calculation into frequency domain calculation,reducing computation time costs.Secondly,Edge Boxes is introduced to estimate the aspect ratio of the target scale,and the results obtained from the fixed aspect ratio of the scale pyramid method are linearly fused with it.The target scale information obtained in this way is different from the fixed aspect ratio obtained by traditional correlation filtering methods and has a more robust scale information.(2)A correlation filter tracking method based on Hilbert-Schmidt independence criterion learning spatial regularization is proposed.Firstly,the high-confidence spatial weight coefficients learned by the filter encourage the filter to focus on more reliable foreground regions.Secondly,the alternating direction multiplier method is used to optimize the proposed model,and the subproblems of the model have closed-form solutions.Thirdly,the model update scheme utilizes the Hilbert-Schmidt independence criterion and the peak-to-sidelobe ratio to obtain the confidence of the target bounding box,reducing the impact of background noise on the UAV target.