Detecting Persons in River Quarry Operations: Aerial Surveillance for Flood Pre-Disaster Monitoring

During the flood disaster preparedness procedure, warning dissemination followed by response decision-making and implementation is usually time-consuming for disaster preparedness teams. Some inaccessible areas that are also prone to flooding is a crucial challenge for the disaster preparedness team in disseminating the early warnings, which may result in endangering the lives of people living in those areas. Moreover, the disaster preparedness team is dedicated to overseeing and managing river quarrying activities, as these are areas where workers or individuals are at high risk of flooding disasters. In this study, an Unmanned Aerial Vehicle (UAV) with Person Detection System that can assist the disaster preparedness team for flood pre-disaster monitoring was introduced. The study investigated the performance of advanced deep learning models, particularly on You Only Look Once, for accurate person detection utilizing UAV imagery focusing in a river-quarry setting. A novel approach to data collection was also presented, specialized datasets were created for capturing river quarry-specific scenarios and used this in facilitating realistic model training. Two models were evaluated in this study, the YOLOv5 model and YOLOv8 model, through their precision, recall, F1-score, and mAP performance. The trained YOLOv5 model showed better performance with 97.70% precision, 94.10% recall, 95.10% F1-Score, and 64.50% mAP than the YOLOv8 model. The significant difference of 2.00% for precision, 3.80% for recall, 3.00% for F1-score, and 0.40% for mAP, between YOLOv5 model and YOLOv8 model, provided beneficial findings for optimizing model selection for person detection using UAV imagery considering other areas needed for pre-disaster monitoring.