[Paper Review] RoadAtlas: Intelligent Platform for Automated Road Defect Detection and Asset Management

Zhuoxiao Chen, Yiyun Zhang+4 more|arXiv (Cornell University)|Sep 8, 2021

Infrastructure Maintenance and Monitoring4 citations

TL;DR

RoadAtlas is an end-to-end AI-powered platform for automated road defect detection and road marking parsing, integrating deep learning models (crack detector and segmenter) with a web-based dashboard and MongoDB backend. It enables real-time, high-accuracy identification of road anomalies—such as cracks and markings—reducing manual inspection costs by up to $400K per survey while improving data timeliness and decision-making for local governments.

ABSTRACT

With the rapid development of intelligent detection algorithms based on deep learning, much progress has been made in automatic road defect recognition and road marking parsing. This can effectively address the issue of an expensive and time-consuming process for professional inspectors to review the street manually. Towards this goal, we present RoadAtlas, a novel end-to-end integrated system that can support 1) road defect detection, 2) road marking parsing, 3) a web-based dashboard for presenting and inputting data by users, and 4) a backend containing a well-structured database and developed APIs.

Motivation & Objective

• Address the high cost and time consumption of manual road condition surveys, which can exceed $400K per survey.
• Reduce reliance on expert inspectors by automating defect and marking detection using deep learning.
• Develop an integrated system that supports real-time data processing, visualization, and reporting for road asset management.
• Enable local governments to monitor road conditions efficiently and proactively plan repairs using predictive analytics.
• Create a scalable, adaptable platform capable of recognizing new defect types and integrating new data over time.

Proposed method

• Deploy a web-based dashboard using Flask and MongoDB to enable user interaction, data validation, and reporting.
• Implement a two-stage deep learning pipeline: first, a crack detector based on YOLO-inspired architecture to localize defects using bounding boxes.
• Second, a U-Net-based crack segmenter trained on pixel-level annotations to produce precise semantic segmentation masks.
• Apply image anonymization to protect privacy by blurring license plates before processing.
• Convert irregular polygon labels into minimal bounding boxes for efficient training of the crack detector.
• Use bird’s-eye view transformation and low-level contour detection to refine road marking segmentation results and improve accuracy.

Figure 1. The overall framework of the RoadAtlas.

Experimental results

Research questions

• RQ1 Can an end-to-end system automate road defect detection and road marking parsing with high accuracy and real-time performance?
• RQ2 How can deep learning models be integrated with a scalable backend and interactive dashboard to support practical road asset management?
• RQ3 To what extent can automated detection reduce the cost and time of manual road surveys compared to traditional methods?
• RQ4 Can the system be extended to detect new types of road defects with minimal retraining or data reconfiguration?
• RQ5 How effective is the combination of bounding box detection and pixel-level segmentation in accurately identifying diverse road defects?

Key findings

• RoadAtlas reduces the cost and time of road condition surveys by automating defect detection, potentially saving up to $400K per survey compared to manual inspection.
• The system achieves high-precision detection of road defects, including Kerb_Cracking, Road_Crocodile, Road_Longitudinal, Road_Transverse, Road_Block, and Sealed_Crack, using a two-stage deep learning pipeline.
• Pixel-level segmentation via the U-Net-based crack segmenter enables accurate localization of defects, with visual outputs mapping predictions back to original images for validation.
• The integration of bird’s-eye view transformation and low-level contour detection improves the accuracy of road marking parsing by aligning predictions with geometric regularities.
• The platform supports real-time data processing and interactive validation through a web dashboard, allowing users to inspect, verify, and export defect and marking data in CSV or JSON format.
• The system is scalable and extensible, capable of adapting to new defect types and new data inputs through modular API design and a well-structured database.

Figure 2. Dashboard (top) and Road Defects View (bottom).