Performance Comparison between OpenPose and TRT_Pose for Self-Practice Yoga on Embedded GPU Platform
Cheng-Liang Shih, Wan-Chia Huang, Irin Tri Anggraini, Yanqi Xiao, Nobuo Funabiki, Chih‐Peng Fan · 2023
In order to enhance the performance of human pose estimation applications on embedded systems, accurately identifying human keypoints and interpreting pose-related information are crucial. In this study, two different pose estimation methods, namely "Openpose" and "TRT_pose", are applied, and their performances are compared by realizing on an embedded platform. The results obtained from these two pose estimation methods are evaluated by analyzing the positional differences of human joint keypoints with Euclidean distance. The study aims to provide insights into improving human pose estimation for embedded systems and facilitating the correct interpretation of human body poses. Additionally, a Yoga self-training system is utilized as the target, and Yoga movements are used as the comparison data.