Search and rescue (SAR) operations require rapid survivor detection, as the chance of locating survivors falls sharply with time. UAVs can quickly survey large, inaccessible disaster areas and provide real-time situational awareness while reducing risks to rescue personnel. However, visible-light imagery becomes unreliable in rain, fog, haze, and darkness, causing missed detections and delayed rescue.
This project proposes a UAV-assisted SAR framework combining enhanced RGB and thermal infrared imagery for robust human detection. In daytime, a unified enhancement model reduces weather-related degradation in RGB images before a YOLO-based detector identifies human targets. At night or in extremely low light, thermal infrared imagery detects survivors through heat signatures, allowing the system to adapt to the operating environment within one unified pipeline. The framework will be evaluated on aerial datasets covering varied weather and illumination conditions using mAP@0.5, mAP@[0.5:0.95], inference speed, localisation accuracy, and robustness across RGB and thermal modalities. The goal is a practical, low-cost, portable decision-support system that improves survivor detection and reduces SAR response times.
Project Team:
Mr. Vinuka Wijesinghe,
Mr. Prabath Samarasinghe,
Mr. Malaka Gunawardana
Mr. Pamudu Lokuhewa
Supervisor: Prof. Dulani Meedeniya