Three-Dimensional Human Hand Pose Estimation Based on Finger-Point Reinforcement and Multi-Level Feature Fusion

ACTA OPTICA SINICA(2022)

引用 0|浏览0
暂无评分
摘要
The existing three-dimensional (3D) human hand pose estimation algorithms do not fully exploit the characteristics of fingers and the key features. To solve this problem, a finger-point reinforcement (FPR) strategy and a multi-layer fusion squeeze and excitation (MFSE) block are proposed. The FPR strategy highlights the role of the finger position points in the human hand point cloud, strengthens the attention of network feature extraction layers to the finger position points in the point cloud, and improves the regression accuracy of the finger joint points. The MFSE block improves the ability of the layered network to extract and express local features. This module realizes the fusion and weight distribution of different levels of features between the layered networks, thereby enhancing the robustness of the model and the accuracy of human hand pose estimation. Experiments on two public benchmark datasets, MSRA and ICVL, verify that the proposed algorithm can achieve high-precision 3D human hand pose estimation.
更多
查看译文
关键词
machine vision,three-dimensional point cloud,deep learning,attention mechanism,hand pose estimation
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要