YOLOv8: An Integrated Framework for Automated Road Damage Quantification and Prioritization

Abstract

The bad state of roads is still a big problem in India, causing accidents, damaging vehicles, and creating delays that impact everyday life and the economy. One of the main problems that officials face is not having up-to-date and specific information about where repairs are needed, which makes it hard to decide which areas to fix first. Most current complaint systems use only text for input, don\'t have good ways to check if the information is correct, and don\'t give much helpful information to the public. To fix these issues, this project presents Street Info Hub, a web platform designed for citizens. It lets users report damaged roads by sharing pictures and providing location details through GPS. The main part of the system is a combined artificial intelligence process. A fine-tuned MobileNetV3 Small model first checks whether the uploaded image contains a road surface. Once validated, a YOLOv8n model trained to detect four types of road damage identifies and marks defects. The severity of each issue is calculated based on damage type, bounding-box area, and image dimensions. All verified reports are stored in a cloud-based MongoDB database along with detailed metadata. The platform also includes tools for analyzing damage patterns and supports route planning by displaying reported potholes on navigation paths using OpenRouteService and Leaflet maps. A dedicated dashboard enables users to monitor their submissions. After training, the YOLOv8n model achieved a mAP@50 score of 46.7% and a precision of 54.7%, showing reliable performance. Overall, the system demonstrates how citizen participation and computer vision can provide real-time data for faster and better road maintenance decisions in practical real-world deployment scenarios today.

Introduction

India’s roads often suffer from potholes, cracks, and other damage that remain unreported or unfixed due to a lack of timely and reliable information. Traditional inspection methods are slow and resource-heavy, while citizen reports are often incomplete. To solve this, Street Info Hub allows users to upload images with location data to report road issues. An AI system then analyzes these images, identifies road damage, and assigns severity scores, which are stored in a database.

The system is built as a modular web platform with three main components: a React frontend, a Node.js backend, and a FastAPI-based machine learning service. It uses a two-stage deep learning pipeline: MobileNetV3 first filters whether an image contains a road, and YOLOv8 then detects and classifies specific road damages such as potholes and cracks.

The literature survey shows that existing systems either focus on detection accuracy, citizen reporting, or mapping, but rarely integrate all three effectively. Many also struggle with real-time performance, usability, or combining AI with geospatial awareness.

The proposed system improves on these gaps by combining AI detection, citizen participation, real-time processing, and route-based hazard visualization. It also includes dashboards, maps, and reporting tools to improve public awareness and infrastructure planning.

In implementation, a dataset of about 47,000 images is used. The system runs efficiently, processing reports in a few seconds while maintaining reasonable detection accuracy. YOLOv8 achieves moderate performance but may struggle with small cracks and uneven datasets.

Conclusion

This project presents Street Info Hub, a web platform focused on helping citizens report road issues. It uses modern web tools along with deep learning techniques and is built on a scalable microservice architecture. The system has a React front end, a Node.js back end, and an ML service made with FastAPI, which lets users send reports in real time, check their accuracy, and view them visually. A big part of this work involves a two-step detection process. First, MobileNetV3 is used for a quick check, and then YOLOv8n is used to detect any damage. This method makes computing faster without lowering performance, and it works well to block out non-road data, correctly finding different kinds of damage and judging how serious each is. The platform also has safe login methods and offers tools such as analytics and visual maps that highlight dangerous areas on routes, making it simpler to use. The system shows that using automated methods led by citizens to check on infrastructure works really well, saves a lot of time, and can get bigger as more people join, making it a strong beginning for smart city projects. Future efforts will focus on improving the system\'s accuracy, its ability to handle larger tasks, and its overall performance. Making the dataset better and training more detailed YOLO models can help improve how well they detect objects. Building an administrative dashboard for officials and a mobile app can make the system more helpful and simpler to use. A big improvement is using LiDAR data along with RGB images to help analyze road damage in three dimensions. This would let us measure depth, volume, and how bad the structure is, which helps in getting the right amount of materials and doing a better check on the infrastructure. Other features such as deleting duplicate reports, sending alerts, and looking at data over time can help make the platform even more useful. This makes it a complete and intelligent system for managing large infrastructure projects.

References

[1] Nilima Pagar et al., Road Damage Detection and Reporting System Using Fully Connected CNN, International Advanced Research Journal in Science, Engineering and Technology (IARJSET), Vol. 11, Issue 4, April 2024. [2] Adi, T. J. W., Suprobo, P., &Waliulu, Y. E. P. R. (2024). iRodd (intelligent-road damage detection) for real-time infrastructure preservation in detection, classification, calculation, and visualization. Journal of Infrastructure, Policy and Development, 8(11), 6162. [3] Dasari, D., Vijaya Lakshmi, D. N. V. S., & Ashok, T. (2024). Real-time detection of road damages using YOLOv8: An innovative deep learning approach. 11(12), h459–h465. ISSN: 2349-5162Journal of Emerging Technologies and Innovative Research (JETIR). [4] Shim, S., Kim, J., Lee, S.-W., & Cho, G.-C. (2022). Road damage detection using super-resolution and semi-supervised learning with generative adversarial network. Automation in Construction, 135, 104139. [5] React Documentation. Available: https://react.dev/ [6] Node.js Documentation. Available: https://nodejs.org/ [7] FastAPI Documentation. Available: https://fastapi.tiangolo.com/ [8] MongoDB Inc., “MongoDB Documentation,” Available: https://www.mongodb.com/docs/ [9] OpenRouteService API Documentation. Available: https://openrouteservice.org/ [10] Leaflet JavaScript Library. Available: https://leafletjs.com/ [11] OpenStreetMap Contributors, “OpenStreetMap: A Collaborative Project to Create a Free Editable Map,” Available: https://www.openstreetmap.org

Copyright

Copyright © 2026 Ram Krishna Verma, Archit Patle, Abhijit Routray, Aalaukik Dole, Ashwini Bajanghate, Anushka Gajbhiye. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.