3D Hand Pose Estimation
Overview
The field estimates a hand’s 3D articulation, shape, and sometimes motion in camera or world coordinates, supporting applications such as interaction analysis, sign-language capture, and robot teleoperation. Common approaches combine image or depth features with parametric hand models, transformers, geometric and kinematic constraints, and temporal or generative motion priors; some also fuse multiple views, inertial sensors, or electromyography. Evaluation spans benchmarks such as FreiHAND, InterHand2.6M, DexYCB, HO3D, HOT3D, and ARCTIC, covering single-image pose, egocentric video, and hand-object interaction. Persistent challenges include occlusion and hands leaving the view, reliable metric scale and camera-motion recovery, and generalization from limited annotations to varied real-world settings.
Latest papers
69 papers16 papers in the last four weeks, up 167% on the four weeks before. 0.1% of all new papers.