Positional accuracy analysis of YOLO personal mobility detection

As personal mobility (PM) becomes increasingly prevalent in urban environments, the precise detection and monitoring of PM is crucial for urban aesthetics and safety. Therefore, in this study, the YOLO algorithm, renowned for its efficiency and effectiveness in object detection tasks, is employed to detect PM from UAV orthophotos. Additionally, the positional accuracy of the detected PM is evaluated using ground-truth data. Given that PM is relatively small compared to other urban objects, the feasibility of detecting and precisely locating PM was analyzed. The assumption that the centroid of the bounding boxes detected by the YOLO algorithm adequately represents the position of the detected objects was also verified. Aerial photos were collected at a 50 m altitude with an RTK UAV over three study sites. Each site contains approximately 15 PM, and their centroid coordinates were investigated through VRS GNSS surveying. For accurate geo-rectification of raw UAV images, five ground control points were installed on each site, and their coordinates were surveyed. The PM detection results showed that the positional accuracy of the PM centers had an error of 13.95 cm. Finally, it was confirmed that UAV photogrammetry and machine learning applications are effective methods for the precise detection and monitoring of PM in urban environments.