Multistage 3D Hand Pose Estimation Algorithm Based on Skeleton Points

Wenqiang Wang,Lihua Tian,Chen Li
DOI: https://doi.org/10.1109/icct56141.2022.10073350
2022-01-01
Abstract:Multi-stage 3D hand pose estimation based on RGB images is a popular research direction in the field of hand pose estimation due to its high accuracy and wide application scenarios. However, there are certain problems and difficulties in this task. Accordingly, we design a multi-stage 3D hand pose estimation network based on skeleton points. Firstly, the joint points that cannot be detected due to occlusion in the 2D hand pose estimation stage are corrected, and a time-series dilated convolutional network is constructed to return the 2D hand pose coordinates to 3D in order to solve the mapping ambiguity problem. Meanwhile, a new loss function is proposed to constrain the generated 3D hand pose coordinates. We validate the network on public datasets, demonstrating the effectiveness and state-of-the-art of our work.
What problem does this paper attempt to address?