| Object tracking is a popular topic of research in computer vision,primarily involving cross-cutting domains such as image processing and pattern recognition,which typically refers to the detection of target position and scale changes in video sequences in order to analyze and understand target behavior.With the development of science and technology and growth in the needs of life,it has a wide application outlook in the areas of military defense,smart city,autopilot,etc.The correlation-filtering based tracking algorithm has exceptional performance,which stands out amongst many tracking algorithms with different concepts and is progressively favored by experts and researchers alike.However,factors such as shape variation,rapid movement,illumination and occlusion still bring great challenges to the research of the algorithm,thus,devising an object tracking algorithm with higher robustness and accuracy is still of broad research importance and application value.This paper analyzes the advantages and disadvantages of the existing tracking algorithm,selects frameworks based on the correlation filtering tracking algorithm,makes a deep research on the problems of object deformation,background blurring,fast motion and occlusion that are easy to appear in video sequences,and proposes some corresponding improvement methods.Finally,the validity of the methods is verified by experiments.The research achievements are divided into three parts as follows:(1)To address the problem that the correlation filtering algorithm cannot perceive the changes in the aspect ratio of the moving targets and is susceptible to tracking failure due to the complex environment,this paper proposes a spatial-temporal regularized correlation filtering algorithm with adaptive aspect ratio.Firstly,the average peak-to-correlation energy(APCE)and peak score are used as references to weigh and fuse each feature response map to achieve accurate results.Additionally,a set of novel one-dimensional boundary filters are presented,integrating near-orthogonality and spatial regularization.These filters can adaptively detect changes in the target scale and aspect ratio by precisely locating the boundaries of the target’s bounding box.Moreover,spatial regularization effectively mitigates the negative impact of the boundary effect on boundary filters.Finally,the learning rate of each boundary filter is adjusted separately according to the peak-to-sidelobe ratio(PSR)to prevent the model from degradation.Experiments show that the proposed algorithm exhibits desirable tracking results and achieves better results compared to other state-of-the-art algorithms for each challenge attribute.(2)In view of the problem that correlation filtering algorithm uses a fixed learning rate for template updating,which makes the model lose accurate template information and makes it difficult to cope with complex scenes such as fast motion,this paper,therefore,proposes a background-aware correlation filtering algorithm based on adaptive updating.First,velocity is used to reflect the change in the target’s background,and the change of the target itself is reflected by the response variation of two frames,the learning rate is obtained by combining the above two parameters to realize an adaptive template updating in the tracking process.In addition,a tree-type scale detection method is designed to improve the accuracy of the algorithm.The scale variation of the target is detected step by step with two scale pools,which have different interval ranges.Experiments show that the proposed algorithm exhibits strong competitiveness in terms of precision and success rate,and has good tracking performance and robustness.(3)Based on the Flask development framework,a Web application system is developed integrating the object tracking algorithm designed and implemented in this paper.Users can interact autonomously on the front-end interface of the system,view real-time images captured by the camera or manually imported video sequences,select the targets in the images,perform object tracking tasks,and view the tracking results.The back-end of the system monitors user operations in real-time,invokes the algorithm to track the selected target,and generates and saves result data and log files.After the design,development,and testing process,the analysis of the system operating results verifies the stability and feasibility of the developed program and the practical significance of the algorithm in this paper. |
