A 3D hand pose estimation architecture based on depth camera

Abstract

Considering the problem of the inability to obtain accurate depth information in 3D pose estimation, this research attempts to use a depth camera to obtain accurate depth information to solve this problem and achieve good results. In the process of research, it is found that the general object detection and evaluation method is not accurate enough under the framework proposed in this paper, so this research proposes an evaluation method suitable for this framework. A standardizer is also designed to optimize the detection effect while achieving efficient tracking objects. Ultimately, inference time is reduced by 35%. The implementation of this research architecture is open-sourced at https://github.com/DumbZarro/BuddHand.