🚑 AI-based system for automatic detection of victim injuries and assessment of consciousness level from visual data.
This repository provides the trained model designed for Search and Rescue (SAR), emergency response, and robotic perception scenarios.
Victim Injury & Consciousness Detection is a deep learning-based solution that analyzes visual data to support emergency response teams and autonomous robotic systems by:
- 🔍 Detecting visible injuries in victims
- 🧠 Assessing the level of consciousness
- 🌍 Supporting diverse real-world emergency scenarios
- 🤖 Enabling integration with autonomous rescue robots and perception pipelines
✅ Injury detection from RGB imagery
✅ Consciousness state estimation
✅ Trained with real-world emergency scenario imagery
✅ Dataset enhanced using AI-generated data
✅ Diversity-aware dataset covering:
- Ethnic diversity
- Age variability
- Gender diversity
This repository contains:
- 📦 Trained deep learning model
- ⚙️ Configuration files
- 📊 Training metadata
- 📄 Documentation for deployment and integration
The model has been optimized for perception tasks in Search and Rescue robotics and emergency assistance systems.
The training dataset was developed through a hybrid strategy:
- Images collected from realistic emergency and rescue scenarios
- Focused on authentic victim conditions and environmental variability
To improve generalization and fairness, AI-based generation techniques were used to:
- Increase scenario diversity
- Balance demographic representation
- Simulate rare or difficult rescue conditions
- Expand injury variability and victim states
The trained model and associated resources are publicly available via Zenodo:
👉 https://zenodo.org/records/18554108