Visual Tracking Algorithm Based On Target Loss Discriminating Mechanism And Its Application

Visual tracking algorithm can accurately,stably and quickly identify the target and locate the target position,which is very important for improving the overall performance of the visual tracking system.However,the practice shows that the visual tracking algorithm is easy to lose the target in complex scenes such as background clutter,similar object interference,target occlusion,target out of the camera's view,etc.For this reason,aiming at the problem that the vision tracking algorithm can't find the target after losing the target,this paper proposes a vision tracking algorithm(YOLORTM)based on the target loss discrimination mechanism,and verifies the feasibility of the algorithm on the vision tracking system of mobile robot.(1)In order to achieve accurate tracking in complex scenes,especially when the target is seriously occluded or out of the camera's view,a vision tracking algorithm(YOLO-RTM)is designed based on the target loss discrimination mechanism.First of all,the YOLO-RTM algorithm detects targets in the first frame of the video through an incremental improvement of real-time object detection algorithm(YOLOv3).Then the real-time learning multi-domain convolutional neural networks for visual tracking algorithm(RT-MDNet)is used to predict the change of the target frame by frame.Finally,the Io U(Intersection over Union)between the detection algorithm and the tracking algorithm is calculated.According to the comparison result of the Io U and the preset threshold value,it is judged whether the target is lost and determined the update mode of the model.Experiments prove that the YOLO-RTM algorithm is superior to the RT-MDNet and MDNet(Learning multi-domain convolutional neural networks for visual tracking)algorithms based on deep learning models,as well as the traditional correlation filter algorithm KCF(High-speed tracking with kernelized correlation filters)and f DSST(Fast discriminative scale space tracking)algorithm.To a certain extent,it solves the problem of finding the target again and continuing to track the target after the target is lost.(2)According to the tracking results of YOLO-RTM algorithm and the motion characteristics of mobile robot,this paper designs a vision tracking system based on turnlebot2 mobile robot.In the motion control strategy of mobile robot,when the vision tracking algorithm determines the target is not lost,the designed control strategy is used to control the robot motion;if the vision tracking algorithm determines the target is lost,the control parameters of the previous frame before the target loss are used to control the robot motion.The experimental results show that the mobile robot vision tracking system designed in this paper can track the target's motion trajectory stably and accurately after applying the YOLO-RTM algorithm.In this paper,the visual tracking algorithm is difficult to retrieve the target automatically after the target is lost.Therefore,we deeply study and analyze the visual tracking algorithm and target detection algorithm,the visual tracking algorithm YOLORTM based on the target loss discriminating mechanism is proposed.And through the comparison test verifies that the tracking accuracy of the YOLO-RTM algorithm is better than the commonly used tracking algorithms;the YOLO-RTM algorithm is successfully applied to the vision tracking system of ROS mobile robot,which verifies the practicability of the algorithm.