Human Pose Detection for Robotic-Assisted and Rehabilitation Environments

Assistance and rehabilitation robotic platforms must have precise sensory systems for human–robot interaction. Therefore, human pose estimation is a current topic of research, especially the safety collaboration evaluation biomarkers. Within this field low-cost marker-less estimators OpenPose Detectron 2 has received much attention their diversity applications, such as surveillance, sports, videogames, assessment in motor rehabilitation. This work aimed to evaluate compare angles elbow shoulder joints estimated by during four typical upper-limb exercises: side flexion, extension, abduction. A setup two Kinect RGBD cameras was used obtain ground truth joint skeleton estimations different exercises. Finally, we provided numerical comparison (RMSE MAE) among angle measurements obtained with OpenPose, 2, truth. The results showed how outperforms these types applications.