Pose-Guided Hierarchical Graph Reasoning for 3-D Hand Pose Estimation From a Single Depth Image.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• 3D hand pose estimation from depth images is crucial for human-computer interaction.
• Existing methods struggle with self-occlusion and finger similarity in complex scenes.

Purpose of the Study:

• To improve the robustness and accuracy of 3D hand pose estimation.
• To explicitly model correlations between different hand parts to address ambiguities.

Main Methods:

• Propose a pose-guided hierarchical graph convolution (PHG) module integrated into a pixelwise regression framework.
• PHG module extracts hierarchical node features guided by hand pose.
• Utilizes graph convolution for hierarchical message passing based on hand structure.
• Generates dynamic convolution kernels for structure-aware feature maps.

Main Results:

• Achieved state-of-the-art or comparable performance on five benchmark datasets (HANDS 2019, HANDS 2017, NYU, ICVL, MSRA).
• Demonstrated enhanced feature maps by exploring complex dependencies between hand parts.

Conclusions:

• The proposed PHG module effectively enhances 3D hand pose estimation by leveraging multipart context and hand structure constraints.
• This approach offers a robust solution for complex scenarios with self-occlusion and high self-similarity.