| Abstract |
More than 2,500 fatal traffic accidents occur annually in Japan, and half of them occur at intersections. Preventing accidents in environments such as intersections where drivers and pedestrians are prone to recognition errors is a significant challenge in achieving safe and secure smart cities. Existing research proposes systems that utilize UAV equipped with cameras to recognize and track vehicles, but these systems struggle to recognize targets in low-light conditions such as at night. On the other hand, other studies are focusing on the use of 3D LiDAR which can observe the three-dimensional structure of real-world environments in detail. However, when the 3D LiDAR is fixed to the ground or a vehicle, there may be many locations where the 3D LiDAR cannot observe correctly because the infrared lasers are often blocked by obstacles or vehicles on the ground. Therefore, in this study, we propose a system that a UAV equipped with a 3D LiDAR flies overhead to observe the whole target area on the ground. By having the UAV observe the ground, the system enables to detect and identify vehicles and pedestrians at intersections without blind spots. The system processes from the data acquisition to the data analysis on an edge device installed on the UAV to enable real-time feedback to the real-world environment. Moreover, the system utilizes not only the 3D LiDAR but also an IMU and GPS receiver to achieve high-precision self-positioning of the UAV and 3D map generation of the target environment. © 2025 IEEE. |
