Research On Recognition And Tracking Of High-Speed Moving Objects Under Binocular Vision

Robot vision system is a more important subsystem in the bionic robot system,the vision system is mainly used for bionic robot information perception of the external environment,can achieve the target of interest to identify and track.In this paper,the identification and tracking of high-speed moving objects is realized under binocular vision,and the system relies on the self-made binocular visual acquisition device,takes the high-speed moving table tennis as the experimental object,analyzes the problems in the actual scene accordingly,and then improves the corresponding improvement from the recognition tracking algorithm,and test its function.The work is as below:（1）Self-made high-speed binocular visual image acquisition device,combined with multi-threading technology to complete the image data processing.Zhang Zhengyou calibration method is used to calibrate the self-made binocular visual acquisition device to obtain the internal and external parameters of the binocular visual acquisition device.（2）In view of the problem that the traditional frame difference method is prone to "hollow" phenomenon,resulting in incomplete detection target contour,the paper proposes a target detection algorithm to improve the three-frame difference method.Since the high-speed moving target will produce certain smear problems during detection,this paper also proposes a center extraction method that combines low-order moment and ellipse fitting to reduce the center position extraction error caused by smearing.For faster access to the position of the target of the movement,this paper proposes a method that combines the idea of dividing the ROI region and the state machine.（3）Through the analysis of KCF tracking algorithm and Kalman filtering algorithm,in view of the shortcomings of KCF’s easy jitter and no automatic detection of occlusion,this paper introduces the SIFT feature into KCF to achieve automatic detection when the moving target is occluded,and proposes an improved algorithm for moving target tracking combining KCF tracking with SIFT feature and Kalman filtering.The algorithm verification shows that the tracking improvement algorithm can continuously and stably track the target when the moving target is occluded.（4）Taking table tennis as the experimental object,this paper designs corresponding experiments to verify the accuracy and wholeness of the improved method in goal identification and the robustness and real-time performance of the target tracking algorithm with or without occlusion,and analyzes the experimental results.The method can realize more precision,integrity and robustness for moving target recognition and tracking in binocular vision system,It contented the demand of real-time.