Building Motion Generation for Sign Language AI

In this demo, Mugi presents an important update to her project by transitioning from a single-camera setup to a dual-camera solution. By introducing a second viewpoint, the system can capture motion and spatial information more accurately, reducing ambiguity and improving overall reliability.

In this session, Munkhchimeg Sergelen talks about how avatars are created for SignMate and why the underlying motion data must be extremely clean. ASL relies on precise handshapes and movement paths, so noisy or unrefined tracking can quickly break the accuracy of the sign.

In this demo, Mugi continues her exploration of motion-capture tools by testing Plask and FreeMoCap to understand how well they can track ASL gestures. She compares their strengths, limitations, and the challenges they present when used for sign-language animation.

In this demo, Mugi walks through her early experiments in capturing and generating ASL motion for SignMate. She explores several prototype approaches, including MediaPipe pose tracking, RPM, and ASL Mocap datasets, evaluating how each tool handles the complexity of sign-language gestures.