REACH: Hand Pose Estimation from Room Corners
Researchers have developed REACH-Net, a novel 3D hand pose estimation system capable of accurately tracking hand shape and pose from fixed cameras in room corners. The system is designed to work with extremely low-resolution and occluded views by leveraging hand-body coordination and temporal progression. To train and evaluate REACH-Net, a new large-scale dataset called REACH was created, featuring 50 participants engaged in daily activities, with hand data captured via concealed chest cameras. AI
IMPACT Enables more robust 3D hand tracking in challenging, real-world environments for applications like human behavior analysis.