3D human skeleton keypoint detection using RGB and depth image

作     者：Jeong, Jeongseok Park, Byeongjun Yoon, Kyoungro 

作者机构：Dept. of Smart ICT Convergence Engineering Konkuk University Korea Republic of Dept. of Computer Science and Engineering Konkuk University Korea Republic of 

出 版 物：《Transactions of the Korean Institute of Electrical Engineers》 (Trans. Korean Inst. Electr. Eng.)

年 卷 期：2021年第70卷第9期

页      面：1354-1361页

核心收录：

主　　题：Cameras 

摘      要：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.