Research / arXiv / Jul 17, 2026
EgoExoMoCap: Distributed Ego-Exo Human Motion Capture
EgoExoMoCap combines egocentric and exocentric signals from people wearing head-mounted devices to reconstruct real-world human motion.
EgoExoMoCap studies distributed human-motion capture using head-mounted devices worn by two or more people. Each wearer contributes first-person observations while also appearing in the other participants' views.
The framework jointly uses egocentric body information and exocentric observations of surrounding people. Head tracking, optional wrist tracking, and context-aware image features contribute to global 3D motion estimation.
The method uses DINOv3-based visual features to handle noise and occlusion. The paper evaluates the approach on two in-the-wild datasets and reports robust reconstruction across the tested scenarios.
Its distributed design shows how several wearable viewpoints can create complementary evidence for embodied interaction and group motion, while keeping capture tied to natural movement in real environments.
The cited arXiv paper contains the full signal model, reconstruction method, dataset description, experiments, and project resources.