Real-time Human Motion Estimation for Human Robot Collaboration

In the process of human robot collaboration, safety is of vital importance, especially when the workspaces of human and robot are intersected, and collisions between them should be avoided. To avoid collision accurately, the motion of people must be in charge in real time, and making a reasonable estimate of human motion, so that the robots can make decisions accordingly, and plan their own motion quickly. This paper presents a framework of real-time motion estimation based on human posture which is based on ROS, firstly, the position of human joints is collected through the Kinect, then the gaussian mixture model (GMM) algorithm and EM algorithm are used to cluster and estimate based on the collected coordinate points, and adding labels to each category, which can help get the sequence of the joint, and realize the function of motion estimation. To guarantee the safety of people, this paper also discusses the motion estimation method of human motion trajectory mutation, which avoids the collision in case of emergency. Finally, the experimental results show that the presented framework of real-time motion estimation can describe the human body's movement accurately and make an accurate prediction, not only ensuring the human security, and it's of great significance in improving the production efficiency.