Automatic estimation of effective road widths to assess disaster response vehicle accessibility in alleys using monocular 3D obstruction detection and road polygons

초록

Rapid access to disaster sites by emergency response vehicles is essential for minimizing casualties and property damage. However, physical constraints such as narrow roads and dynamic obstructions-e.g., illegally parked vehicles-pose significant challenges. While CCTV and dash cameras provide visual information, accurately estimating road width from 2D images remains difficult. Although LiDAR offers high-precision measurements, its cost limits feasibility in large-scale disaster scenarios. This study presents a method for evaluating alley accessibility for disaster response vehicles by estimating effective road width using a monocular camera. The YOLOv5 model is employed to detect ten categories of obstructing objects (e.g., vehicles, poles), while the FCOS3D algorithm estimates their 3D positions and dimensions. Using GIS-based road polygon data, the system calculates the base road width and subtracts the spatial dimensions of detected obstructions to derive the effective road width. For validation, LiDAR point cloud data from disaster-prone alleys in Seoul were used. Bland-Altman analysis showed a mean bias of-0.28 m within the 95% confidence interval. When 3 m and 4 m thresholds were applied, critical error rates-misclassifying Inaccessible or Difficult alleys as Accessible-were only 2.5% and 10.9%, confirming the method's accuracy and conservative reliability for emergency-response planning. The entire evaluation is completed in under 1 s. This approach demonstrates high accuracy across diverse urban settings and obstruction conditions, offering a practical, low-cost alternative to conventional surveys. It enables continuous monitoring of disaster-vulnerable areas and accurate, timely assessments of emergency vehicle access. By capturing dynamic urban conditions, the system supports adaptive response and infrastructure planning, enhancing urban resilience and disaster preparedness.

키워드