Realistic Emergency Video Synthesis

Based on EU Directive 2019/882 (EAA) and the German Accessibility Strengthening Act (BFSG), public safety answering points (PSAPs) in Germany must enable accessible 112 emergency calls by June 28, 2027. Within this framework, the addition of video telephony (Total Conversation service) is gaining significant importance. This will allow dispatchers to perform initial visual assessments of patients (e.g., breathing, skin tone, consciousness) and opens up potential for AI-supported analysis. However, the development of such AI systems is limited by a lack of ethically usable video datasets depicting real medical emergencies.

The REViS research project addresses this bottleneck by developing a generative video model to produce synthetic, physiologically realistic emergency scenes. The generated video sequences are based on structured parameters (e.g., heart and breathing rates, SpO₂ values, camera trajectories) as well as unstructured free-text scene descriptions.The key deliverables include:

- REViS Generator: A modular generator that combines biophysical simulations (e.g., cyanosis-induced skin tone changes, thoracic expansion during respiration) with text-to-video models and style transfers (e.g., smartphone camera artifacts).

- REViS Synthetic Dataset & Model: An annotated synthetic dataset for fine-tuning a specialized video model capable of generating physiologically accurate emergency scenes.

- REViS Demonstrator: A testbed for validating vital signs prediction AI models using established quality metrics.

REViS creates a secure, reproducible, and privacy-compliant training foundation for AI systems in emergency medicine, driving life-saving assistance systems for dispatch centers toward operational readiness.

• Biophysikalische Simulation
• Computer Vision
• Generative KI
• Tele-Notfallmedizin
• Visuelle Vitaldatendiagnostik