Experimental Performance Evaluation of Computer Vision for an Upper Limbs Rehabilitation Robot

Abstract

Rehabilitation robot for upper limbs can help physicians or therapists in conducting repetitive rehabilitation to patients. This robot is constructed by several electromechanical actuators, sensors, controllers, and the main computer. The patient holds a robot gripper in the left and right hands and moves the gripper according to a rehabilitation protocol determined by the physician who is in charge. A computer vision-based position sensor is designed to measure wrist position without using any marker. Moreover, a new algorithm is proposed to filter coordinate values measured by the video camera and extract the position estimate of the wrist. This sensor system estimates positions of the wrist using OpenPose library. The experimental performance was evaluated in terms of precision and speed for several resize values. Experimental results indicate that the pose-based computer vision algorithm can potentially be used for the upper limbs rehabilitation robot. Our finding stimulates further research to find more consistent values of the precision performance index.