Enhancing road safety: A convolutional neural network based approach for road damage detection

Abstract

Road damage poses considerable challenges for both conventional and autonomous vehicles, with obstacles such as potholes, speed bumps, cracks, and manholes increasing the risk of vehicle damage and accidents. For autonomous systems, the ability to detect these hazards in real time is essential to ensure passenger safety and protect vehicle integrity. In this paper, we introduce a comprehensive road damage dataset encompassing these four common types of damage and present the DD-CNN-23Layers model, a convolutional neural network specifically designed for road damage detection and classification. We benchmarked our model against pretrained YOLO models (versions 7 to 10), with the DD-CNN-23Layers model achieving a precision of 91.86% and a mean Average Precision (mAP) of 97.54%, outperforming all compared YOLO models. By utilizing this model, autonomous driving systems can proactively respond to road hazards, improving navigation safety and extending the lifespan of both vehicles and infrastructure.