Vision Based Target Following UGV System Using YOLOv5 and ROS Platform

Abstract

Realizing the autonomous following function of the unmanned ground vehicle is critical for its applications in the industrial, medical, and home service fields. This paper proposes a visual autonomous following unmanned ground vehicle system for indoor environments based on the ROS platform, which combines autonomous navigation and visual following technology, senses unknown environment with RGBD camera and LIDAR, detects tracking targets with the YOLOv5 algorithm, meanwhile integrates the KCF algorithm and pedestrian re-identification technology for target tracking. The unmanned vehicle system can detect user-specific gestures to start and stop, and during the tracking process, the transfer of different states is realized by a state machine, then the motion control and dynamic obstacle avoidance are implemented using simultaneous localization and mapping (SLAM) algorithm. Finally, the above process is validated using a real unmanned vehicle in the indoor environment, and the results show that the autonomous following UGV system can effectively avoid obstacles and track the target to the specified location, which verifies the availability and reliability of the system.