Human Pose Estimation Using GNN

Human Pose Estimation is a method of capturing a collection of coordinates for each joint (arm, head, torso, etc.) that may be used to characterize a person's pose. The initial goal is to create a skeleton-like depiction of a human body, which will then be processed for task-specific applications. The ability to identify and estimate the position of a human body is valuable in a wide range of applications and conditions like action recognition, animation, gaming, and so on. It is a crucial first step toward understanding people through images and media. In this study, graph neural networks were utilised to predict human poses by modelling the human skeleton as an unordered list, greatly enhancing 3D human pose estimation. This paper describes the approach as an efficient way to determine the 3D posture of many persons in a picture. Our model gives a validation accuracy of 92%.