Home // eTELEMED 2021, The Thirteenth International Conference on eHealth, Telemedicine, and Social Medicine // View article
Authors:
Oky Dicky Ardiansyah Prima
Kazuki Hosogoe
Keywords: 3D human pose; body joint; computer vision.
Abstract:
Three-dimensional (3D) human pose estimation has been used in a wide range of fields, including motion analysis in sports and rehabilitation, modeling in Computer Graphics (CG) production for movies and games, and input interfaces. Recently, 3D human pose can be estimated with high accuracy only from a single image using a neural network model. However, depending on the camera's position and shooting angle, some joints may be occluded, thus reducing the accuracy of the overall joint estimation. In this study, we experimentally constructed a neural network model for 3D human pose estimation based on a single image and evaluated the difference in accuracy of the pose estimated by the model constructed for the partial joints of the body and the whole-body joints. The dataset used for the experiment was Human 3.6M and a human pose dataset created from an RGB-D camera for this study. The results confirmed that the model built based on the upper-body joints of the body had higher accuracy than that for the whole-body joints at estimating the posture of the upper body. Finally, we demonstrated that 3D human pose can be used to detect the deterioration in sitting postures, which can suggest that the technology is effective in improving various postures in daily life.
Pages: 1 to 5
Copyright: Copyright (c) IARIA, 2021
Publication date: July 18, 2021
Published in: conference
ISSN: 2308-4359
ISBN: 978-1-61208-872-3
Location: Nice, France
Dates: from July 18, 2021 to July 22, 2021