Multiscale Feature Fusion Network for Monocular Complex Hand Pose Estimation

• DOI: 10.22541/au.169743436.63648882/v1
• Publication Date: 2023-10-16T05:32:46Z
• Authors: Zhi Zhan; Guang Luo
• Keywords: Feature (linguistics); Monocular; RGB color model; Similarity (geometry); Monocular vision

Hand pose estimation based on a single RGB image has low accuracy due to the complexity of pose, local self-similarity finger features, and occlusion. A multiscale feature fusion network (MS-FF) for monocular vision gesture is proposed address this problem. The can take full advantage different channel information enhance important information, it simultaneously extract features from maps resolutions obtain as much detailed deep semantic possible. are merged hand results. InterHand2.6M dataset Rendered Handpose Dataset (RHD) used train MS-FF. Compared with other methods (which estimate interacting poses image), MS-FF obtains smallest average error joints RHD, verifying its effectiveness.