Multi-view 3D Markerless Hand Motion Capture System with Keypoint Triangulation

This paper presents an approach for markerless hand motion capture using a multi-view setup. Our system leverages the flexibility of multiple color cameras mounted on tripods or a rigid frame to accommodate two different capture volumes. An efficient and accurate calibration process is facilitated by an active dual-LED wand. Machine learning technique is used to detect 2D keypoints from each camera image, which are then triangulated to reconstruct 3D joint positions. The system demonstrates the potential for advancing clinical research through its flexible setup and comprehensive capture of both hand and body motion.