Trajectory Planning of Upper Limb Rehabilitation Robot Based on Human Pose Estimation

Abstract

Stroke has become the second leading cause of death in the world, and timely rehabilitation can effectively help patients recover. At present, with the shortage of rehabilitation doctors, using rehabilitation robots to help patients recover has become a more feasible solution. In order to plan a bionic motion trajectory of an upper limb rehabilitation robot more conveniently, a teaching trajectory planning method was proposed based on human pose estimation in this paper. The teaching trajectories were collected by Kinect's depth camera and human bone joints were tracked using deep neural networks OpenPose. The processed trajectories were verified with modeling simulation and robot motion. The planar trajectories were evaluated using the minimum Jerk principle on bio-imitability, the position determination coefficient is more 0.99, the speed determination coefficient is more than 0.94, and the acceleration determination coefficient is more than 0.88. In the case of block, the recognition success rate has increased by more than 73.4% compared with Kinect's bone binding OpenPose algorithm for human bone joint recognition. The bioimitability of the trajectories planned by this method can conveniently and quickly meet the needs of rehabilitation doctors in hospitals to plan the rehabilitation robot trajectory.