Automated Detection of Roadway Obstructions to Assess Vehicle Accessibility Using Aerial and Reference Images

Abstract

In the aftermath of natural disasters, roadway obstructions can hinder access to impacted communities, which can severely impact emergency response and evacuation efforts. Traditional ground-based and aerial reconnaissance methods for obstruction detection are often limited by cost, accessibility, and efficiency. This study introduces a novel framework that compares postdisaster unmanned aerial vehicle (UAV) images with predisaster satellite images to detect and segment roadway obstructions and estimate remaining accessible road width to provide emergency managers with updated status of roadway networks. The approach uses the You Only Look Once version 8 (YOLOv8) algorithm to segment aerial roadways, and then these are compared with predisaster reference images at the same location to identify changes in road conditions, notably reducing false positive results and enhancing detection accuracy. Due to a dearth of availability in training data for UAV-based aerial images of obstructions on roadways, synthetic data are generated through data augmentation techniques to bolster model performance. The developed framework achieved a mean average precision (mAP) of 98.5% (mAP 50), which evaluates detection accuracy at an Intersection Over Union (IoU) threshold of 0.5, and 91.2% (mAP 50–95). Results demonstrated improved prediction accuracy with reference images, achieving a 94.67% success rate compared with 48% without them. The methodology enables precise estimation of road usability for various vehicle types, facilitating efficient route planning and debris clearance. This research advances postdisaster roadway assessment by leveraging UAV and photogrammetry techniques, offering a rapid and accurate solution for postdisaster management, and planning for recovery operations.