Vision-based Moving Target Tracking of Certain Target for Quadruped Robots

摘要

Due to the high flexibility of quadruped robots compared with some traditional robots, it has become an important branch in the field of mobile robot research. Target detection and tracking technology is important for the environment perception and autonomous motion tasks of quadruped robots. To achieve effective tracking of a specific target, the premise is that the target is accurately detected. In this article, a target detection algorithm based on deep learning for quadruped robot person tracking is proposed. After the targets are detected, a simple but accurate selection strategy is used to extract the certain target to be tracked from other targets. The visual distance and angle information from the target is used to realize the autonomous tracking. In the part of experiment, we conducted experiments based on the changeable outdoor environment, which included interference factors such as light changes, people crossing, obstruction by obstacles. During the experiments, the SDUQuad-48 quadruped robot can simultaneously realize the stable tracking of a specific moving target based on our algorithm with a speed of 0.8 m/s.