Design of an edge computing-based motion capture and model-driven system

Motion capture has become a core technology for applications involving virtual digital humans, such as virtual streamers (VTubers). This paper proposes a motion-capture and model-driven system that relies on keypoint detection performed on edge-computing devices. The system implements independent keypoint detection for multiple body parts and parallel real-time processing on edge devices, enabling real-time keypoint detection of the body, hands, and face. By offloading tasks to edge devices, it reduces the resource footprint of motion capture and lowers deployment costs. Experimental results show that employing edge-computing devices can significantly reduce device load and enable deployment across a variety of platforms.