Research On Multi-target Tracking System Of Hexapod Robot Based On Kinect

In recent years,the target tracking technology has received extensive attention from researchers,but there are three key issues that need to be solved urgently in the target tracking technology.One is that the cameras used for video shooting are mostly fixed on a tripod and cannot be moved arbitrarily,resulting in very limited monitoring scenes;The second is that target detection and tracking are mostly applied to wheeled robots.Wheeled robots cannot perform stable shooting on uneven roads.The third is due to factors such as outdoor light intensity,direction changes,shadow occlusion,and complex backgrounds.Target detection and tracking have disadvantages such as low accuracy and weak generalization ability.In response to the above three problems,this article conducts multi-target detection and tracking experiments in indoor halls,outdoor corridors,and gravel roads.The six-legged robot is equipped with a Kinect camera to collect video files.Through the improved YOLO v3 combined with Deep-SORT algorithm Real-time detection and tracking of multiple targets appearing in the video.In order to meet the requirements of real-time and accuracy of multi-target detection and tracking,and to enhance the detection performance of the hexapod robot target detection network in the distance and the ability to obtain detailed features,it is added after the second residual structure in the YOLO v3 network A residual structure containing four residual units is used to extract more target feature information.Due to the introduction of the target tracking algorithm,it has a certain mitigation effect on the phenomenon of target "drop frame" and occlusion in the target detection of the hexapod robot.Aiming at the negative impact of the repeated marking of the area caused by the repeated detection of the target,this article uses Deep-SORT algorithm performs association matching between the two frames before and after the video to achieve stable tracking of the target object.This paper performs multi-target detection and tracking tasks of hexapod robots in indoor halls,outdoor corridors,and gravel road environments.The hexapod robots are equipped with Kinect cameras to collect video files,and the improved YOLO v3 algorithm is used to compare the hexapod robots and hexapod robots in the video.Pedestrians are detected,and the m AP of the recognition of pedestrians and hexapod robots on the test set can reach 94.8% and 92.2%,respectively.After target detection,the Deep-SORT algorithm is introduced to track multiple targets appearing in the video in real time.In an indoor lobby environment,the accuracy of multiple target tracking is 94.6%,and the frame rate is 27 frames per second.In an outdoor environment,more the accuracy of target tracking is 91.5%,and the frame rate is 24 frames per second.The experiment compares and analyzes the accuracy and real-time performance of the algorithm in this paper and the classic tracking algorithm in the multi-target tracking experiment of a hexapod robot.The experimental results prove that the method proposed in this paper effectively solves the accuracy of multi-target detection and tracking in an outdoor environment.Insufficient defects such as low performance and weak generalization ability,and meet the real-time requirements of hexapod robot multi-target tracking.