Accurate 3D hand pose estimation network utilizing joints information

作者：

Highlights：

• We propose a method that concatenating the 2D information and the 3D position of adjacent joints to estimate the depth coordinates of the joints in each stage. The 2D hand joints information is obtained by multi-task method.

• We designed several groups of comparative experiments to prove the effectiveness of our method. Our method also has a good performance when compared with the current state-of-the-art-performing methods.

摘要

•In this paper, a new multi-stage network according to the position of the joints on the finger is proposed to improve the utilization of the information between adjacent joints. The result of each stage is used to the next stage as auxiliary information.•We propose a method that concatenating the 2D information and the 3D position of adjacent joints to estimate the depth coordinates of the joints in each stage. The 2D hand joints information is obtained by multi-task method.•We designed several groups of comparative experiments to prove the effectiveness of our method. Our method also has a good performance when compared with the current state-of-the-art-performing methods.

论文关键词：Hand pose estimation,Deep regression,Multi-stage,2D CNN,Depth image

论文评审过程：Received 20 December 2019, Revised 18 September 2020, Accepted 8 October 2020, Available online 11 October 2020, Version of Record 14 October 2020.

论文官网地址：https://doi.org/10.1016/j.image.2020.116035