3D human skeleton keypoint detection using RGB and depth image

As computing technology advances, tasks those are used to judge human behavior with the eyes are turning into tasks those computers try to judge human behavior through keypoint detection. Accordingly, in this paper, we propose a 3D human skeleton keypoint detection system using RGB and Depth images acquired by Azure Kinect's RGB camera and Depth camera, respectively. The 3D human skeleton keypoint detection system proposed in this paper detects 2D human skeleton keypoints from RGB images, and uses depth value acquired to project the detected 2D human skeleton keypoints onto the depth image. However, when detecting 3D human skeleton keypoints in such method, the human skeleton keypoints are projected onto the surface of human body. To solve this problem, the skeleton keypoints provided by Azure Kinect is used to calibrate the depth value of the extracted keypoints. Copyright © The Korean Institute of Electrical Engineers.