Comparative Analysis of OpenPose, PoseNet, and MoveNet Models for Pose Estimation in Mobile Devices

Pose estimation is a significant strategy that has been actively researched in various fields. For example, the adopted for motion capture moviemaking, and character control video games. It can also be applied to implement user interfaces of mobile devices through human poses. Therefore, this paper compares analyzes four popular pose models, namely, OpenPose, PoseNet, MoveNet Lightning, Thunder, using pre-classified images. The results show Lightning was fastest, OpenPose slowest among models. But only model capable estimating poses multiple persons. accuracies Thunder were 86.2%, 97.6%, 75.1%, 80.6%, respectively.