EgoForce: Forearm-Guided Camera-Space 3D Hand Pose from a Monocular Egocentric Camera

Researchers introduced EgoForce, a monocular egocentric hand-pose framework that recovers absolute 3D hand pose and position from a single head-mounted camera.

It combines forearm-guided representation, an arm-hand transformer, and a closed-form ray-space solver to address depth-scale ambiguity.

The system works across fisheye, perspective, and distorted wide-FOV cameras without needing device-specific retraining.

EgoForce achieved state-of-the-art results on egocentric benchmarks, including up to a 28% reduction in camera-space MPJPE on HOT3D.

The approach could improve hand tracking for AR/VR and telepresence by making robust hand pose estimation easier to deploy on diverse optics.