A Novel YOLOv8-Pose-Based Algorithm for Abnormal Behavior Analysis of Quadruped Robots

In this paper, a novel abnormal behavior analysis algorithm based on YOLOv8 is proposed for the target recognition problem of guide robot dog. The algorithm is based on the human posture estimation model of YOLOv8 for abnormal behavior analysis, and by extracting the coordinate information of the key points of the human body, it can make conditional decisions on abnormal behaviors such as falling forward, backward and sideways, so as to realize the real-time monitoring of the blind people and send out alarms in time. The proposed algorithm ensures the accuracy and speed of the recognition while not taking up a lot of computational resources of the computer. The proposed method is applied to a real quadruped robot platform, and the effectiveness and applicability of the proposed algorithm is verified through a large number of experiments, which in this paper are based on the actual data collected from different human posture pictures.