Fast 3D Hand Pose Estimation for Real-time System
Jae-Hun Song, Suk‐Ju Kang · 2020
In this paper, we propose a novel system that estimates 3D coordinates of the hand joints from a single depth image using a convolutional neural network (CNN). The proposed system consists of the CNN-based feature extraction, and a regression neural network that estimates 3D joint coordinates. The regression network is composed of two branches by dividing a palm and fingers. In the experimental results, the proposed system was compared with other state-of-the-art methods using a public NYU dataset. The proposed system had an average 3D distance error of 9.62mm and inference time of 172 fps. It had the best performance in terms of both accuracy and speed than other recent methods.