AI-Based Drowning Detection in UAV-Assisted Coastal Environment

With the recent advancements in the aviation industry, UAV (Unmanned Aerial Vehicle) technology is being utilized across various fields. Furthermore, it is evolving into an autonomous driving system powered by AI(Artificial Intelligence) and big data. Leveraging these advantages, the current trend in UAV technology is to deploy these vehicles in public areas that are inaccessible to people, particularly for search and rescue operations. In this study, we propose a UAV-based drowning detection method aimed at rescuing individuals who may be at risk of drowning in summer beaches. To address this issue, we employed YOLOv5(You Only Look Once version 5) as the object detection technology. YOLOv5 is known to have low capacity and fast speed. It is designed and used by implementing the same CSPNet-based backbone as Yolov4 in Pytorch. The simulation results indicate that the proposed model can effectively detect three states: drowning person, swimming individuals and people out of the water. Finally, simulation results show that the learned model achieved a detection accuracy of over $90 \%$ on average in sorting three states for several test scenarios.