A New Real-World Dataset for AI-Based Road Damage Detection

01/06/2026

Researchers from the Information Processing and Telecommunications Center (IPTC) at Universidad Politécnica de Madrid, together with collaborators from Sapienza University of Rome and the University of Udine, have developed a new real-world dataset for automated road damage detection. The dataset contains more than 2,000 annotated images collected in urban and rural areas of Italy using low-cost consumer devices, including a GoPro camera and a smartphone.

The dataset contains three key road-surface categories: potholes, cracks, and maintenance holes, including over 4,700 manually verified annotations. Unlike many existing datasets, it was designed to reflect realistic driving conditions, different weather and lighting situations, and cross-device variability, enabling the development of more robust and generalizable artificial intelligence models.

The study also evaluates several state-of-the-art object detection architectures, demonstrating the effectiveness of lightweight AI models for real-time road inspection systems on embedded and edge devices.

This research can support multiple applications,

✅ Smart-city infrastructure monitoring

✅ Predictive Road maintenance,

✅ Autonomous driving systems,

✅ Driver-assistance technologies, and

✅ Low-cost mobile inspection platforms for municipalities and transportation agencies.

Overall, this work highlights the importance of realistic, diverse, and well-annotated datasets for advancing AI-based road inspection systems. By combining real-world acquisition conditions with lightweight detection models, the proposed dataset provides a strong foundation for future research and practical deployment of intelligent road monitoring solutions, contributing to safer transportation networks and more efficient infrastructure maintenance.

Bibliographic reference:

Giordani, E., Arcioni, L., Gil-Martín, M., Foresti, G.L., & Marini, M.R. Real-world road damage dataset with potholes, cracks, and maintenance holes in Sci Rep, art. 16, 15318 (2026). https://doi.org/10.1038/s41598-026-46679-4

This work is the result of a collaboration between Manuel Gil Martín, from the Universidad Politécnica de Madrid, and Marco Raoul Marini, from Sapienza Università di Roma.

Manuel Gil – Martín: GS / ORCID / LinkedIn

Marco Raoul Marini: ORCID / LinkedIn

LinkedIn: https://www.linkedin.com/company/iptc-upm/

For more information: www.iptc.upm.es