3D Human Pose Estimation Using Egocentric Depth Data
Seongmin Baek, Youn-Hee Gil, Yejin Kim · 2024
In this paper, we present a novel approach for 3D human pose estimation using depth data from egocentric viewpoints. Depth data has the advantage that it is less sensitive to color and lighting changes. We acquired depth data streamed from multiple depth cameras attached to a user’s head and calibrated them into a depth map. For joint detection, a ResNet-based network was optimized with the skeletal joints of a Kinect camera. Unlike previous approaches, the proposed approach can track 3D human poses in an egocentric setup with a small dataset.