A Technical Demonstration on Streamlining 3D Motion Production Workflows with a Markerless Motion Capture System

This paper presents an innovative markerless motion capture system to streamline 3D animation workflows using spatial computing and deep learning. We employ off-the-shelf cameras and neural networks to automatically reconstruct 3D joint positions from video. Our lightweight solution retargets motions onto character rigs without manual intervention. With accurate real-time tracking of up to 30 fps, our system has faster processing than the previous method. Our system contributes an easy-to-use, cost-effective motion capture solution by replacing specialized equipment with learned spatial reasoning. This work democratizes immersive content creation by streamlining motion capture based on computer vision and deep learning, helping make high-fidelity capture broadly accessible.