Answers are grounded in this paper and cite their sources.
Rather than treating road distress detection as a single, uniform task, this review considers how inspection needs change between carriageways and sidewalks. It first follows the development of recognition methods, from field surveys and rule-based image processing to handcrafted descriptors, deep feature learning, and real-time detection. Particular attention is given to the YOLO family, including YOLOv5, YOLOv8, YOLOv11, YOLOv12, and YOLOv26. Their feature extraction designs, detection heads, and deployment requirements are discussed alongside published benchmark and road-distress results. The reported evidence shows that a larger network is not necessarily the better choice: designs addressing the visual properties of pavement damage can offer accuracy gains with lower computational demands. Such comparisons must, however, be interpreted in light of the dataset, evaluation protocol, and hardware used. The review draws out the implications for choosing models in practical road inspection.
No takes yet. Share an insight, caveat, or question.
Pan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: