EgoEMS: A High-Fidelity Multimodal Egocentric Dataset for Cognitive Assistance in Emergency Medical Services

EgoEMS is a high-fidelity egocentric, multimodal end-to-end dataset designed to advance AI cognitive assistants for Emergency Medical Services (EMS). It captures ~20+ hours of egocentric (head-mounted) video across 233 simulated emergency scenarios with 62 participants (including 46 EMS professionals) performing realistic, protocol-aligned procedures. Developed with EMS experts and aligned to national standards, EgoEMS emphasizes field realism, team coordination, responder-patient/bystander dynamics, movement, noise, and variable environments while enabling rigorous, reproducible research. EgoEMS provides rich annotations for keysteps (actions) performed by responders for activity recognition, speaker diarized timestamped audio transcripts, bounding boxes with labels for selected medical objects, distance sensor data for CPR ground truth used for quality estimation tasks.