Research on Mobile Robot Autonomous Following System based on ROS

We constructed an autonomous pedestrian target following system based on the ROS (Robot Operating System) platform, including building a mobile robot simulation environment, deep image processing, and differential motion control. And we proposed a deep learning-based re-detection tracking method for pedestrian target tracking, based on loading the YOLOv3-Tiny model and screening the target frame with the highest DIoU value to re-initialize KCF and update the target frame when the KCF algorithm fails to track. We tested the accuracy of the tracking algorithm and the feasibility of implementing it on a mobile robot experimentally. Finally, we designed a pedestrian following an experimental scenario and compared the angular and linear velocity values of the robot at different distances, which demonstrated the stability of the autonomous pedestrian following.