Rapid and Safe Human Detection in Uninhabited Terrains Integrating Formation Flight and Multispectral Imaging

Abstract

Search and rescue (SAR) operations are frequently triggered by hazardous situations, necessitating rapid and accurate detection of individuals in distress. However, environments where such individuals are often found, like forests, present significant challenges for detection. While many studies advanced SAR technologies, including human detection, SAR-specific datasets, and the utilization of unmanned aerial vehicles (UAVs) with capabilities such as collision avoidance and formation flight, it is difficult to find a study which integrates these elements into a single comprehensive approach. In this study, we propose a formation flight system for SAR missions, equipped with peer-to-peer communication for coordination, obstacle identification and avoidance, and RGB and IR cameras to robustly detect individuals. Additionally, we present a novel multispectral dataset for SAR, captured from UAVs over diverse environments, including woodlands, forests, and mountainous regions. We validated that our system could robustly detect individuals while avoiding obstacles during flight in real-world experiments.