Research On Precise Tracking Algorithm Of Spherical Multi-target

In recent years,artificial intelligence has developed rapidly,especially in the field of machine vision.Among them,target tracking is one of its important research branches.It has been widely developed in military,medical,monitoring,human-computer interface,motion analysis and other related fields.It also provides important technical support for event processing,robot navigation and other high-level visual tasks,and also causes in-depth research by researchers in related fields.With the continuous development of target tracking technology,more and more tracking technology can achieve the positioning and tracking of spherical targets.However,due to the complex situation of occlusion and scale change in the tracking process,the traditional tracking technology can not track the spherical targets accurately and stably,so as to achieve the ideal tracking results.Therefore,aiming at these problems,this paper makes a deep research on spherical target tracking,the main research contents are as follows:(1)Spherical single target tracking.On the basis of background sensing filter tracking,a filter combining time regularization and background sensing is proposed.The training samples are extracted from the real background,the classification ability of the filter is enhanced by increasing the training samples,and the time regularization is introduced to construct the target relocation module under occlusion.The alternative direction multiplier method is used to optimize the solution of the target,reduce the computational complexity,and the linear interpolation method is used to update the location and scale of the target.(2)Spherical multitarget tracking.Based on the detection algorithm of YOLOv3,a multi-target tracking algorithm based on YOLOv3 is proposed.The spherical target is detected by YOLOv3 to initialize and create a tracker;the state prediction and covariance prediction generated by the information of the previous frame's bounding box are predicted in the Kalman filter;all prediction results of the Kalman filter are matched by the Hungarian algorithm to remove the matching match whose matching value is less than the threshold value;the bounding box informationdetected by the matched target in the current frame is used to update The new Kalman tracker calculates the Kalman gain,state update and covariance update;outputs the state update value as the tracking frame of the current frame;reinitializes the tracker for the target that does not match in this frame.Through the experimental results and analysis,the spherical target tracking algorithm in this paper has better tracking accuracy than other tracking algorithms in the case of occlusion and scale change,and can track the target stably.