Advancing Human Detection in UAV-Based Search and Rescue Using Transfer Learning Approach

In rapid emergency situations, accurate detection of humans is essential for effective search and rescue operations, especially when using Unmanned Aerial Vehicles (UAVs) for quick response. This paper explores the performance of state-of-the-art You Only Look Once (YOLO) algorithms in identifying individuals in complex disaster scenarios. By utilizing the Combination to Application (C2A) dataset, that provides a diverse range of synthetic images depicting people in challenging environments, including floods, forests, fires, and collapsed buildings. Among the models evaluated, YOLOv11 demonstrated highest accuracy and robustness in detecting people across varied positions and backgrounds. By leveraging transfer learning, this approach achieved a precision of 92%, showcasing significant improvements in human detection. This study also highlights the potential of advanced algorithms like YOLOv11 to enhance the speed and accuracy of victim identification during UAV-assisted search and rescue missions in disaster-stricken areas.