Movement Direction Control for Mobile Robot by Human Position Tracking and Fall Detection Method using OpenPose

Abstract

Tracking human’s movement is a necessary requirement for service robots in places providing the medical care. Therefore, the service robots must firstly recognize the direction and position of the target person moving as well as the distance between the person and robot. To estimate and detect human body motion pattern recognition, there are several approaches method to be being researched and developed. Human-skeleton extraction plays a key component to recognize human motion. This paper proposes a trial method to follow humans and detect human-fall based on walking pattern estimation and humanskeleton data by using OpenPose. In this paper, the authors focus on measuring and calculating the length between the key points of human-skeleton (i.e.head/No.0, right shoulder/No. 2, left shoulder/No. 5, waist /No. 8, left hip/No. 9, and right hip/No.12 of OpenPose) and robot in pixel coordinate for determining the user’s direction when the user moves by using a depth camera. In addition, several individual joint positions are used to detect falling behavior. Robot Operating System (ROS) is used to receive vision messages to control the mobile robot. Furthermore, the methodology and algorithm for human-follow and human-fall detection are validated through the experimental results.