Development of a Low-Cost Hybrid Mobile Mapping System for Automated Pothole Detection and Geometric Quantification Using AI and Deep Learning

• Grant number: 25/27380-5
• Support Opportunities: Scholarships in Brazil - Master
• Start date: June 01, 2026
• End date: July 31, 2027
• Field of knowledge: Physical Sciences and Mathematics - Geosciences - Geodesy
• Principal Investigator: Henrique Cândido de Oliveira
• Grantee: Hasnat Hameed
• Host Institution: Faculdade de Engenharia Civil, Arquitetura e Urbanismo (FEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

Potholes are a recurring and costly problem for road networks, particularly in developing countries where limited budgets and irregular maintenance practices hinder timely repairs. Traditional inspection methods, often based on manual surveys or user complaints, are slow, subjective, and spatially inconsistent. Although high-end Mobile Mapping Systems and LiDAR platforms offer precise measurements, their high cost restricts their adoption by municipalities and small institutions. This research proposes a low-cost hybrid mobile mapping framework that integrates vehicle-based imaging, UAV photogrammetry, and deep learning to support automated pothole detection and geometric quantification. The first stage employs consumer-grade cameras mounted on a terrestrial platform to collect continuous road imagery, which is processed using state-of-the-art AI models (e.g., YOLO and U-Net) to detect potholes under diverse lighting and pavement conditions. In the second stage, a UAV is deployed to the detected locations to generate high-resolution 3D reconstructions using photogrammetric techniques, enabling the extraction of area, depth, and volume metrics. All data are georeferenced using PPK/RTK GNSS solutions and stored in a PostGIS geodatabase for analysis. The system will be validated through comparisons with SLAM-based LiDAR measurements and evaluated using detection accuracy, geometric accuracy, and statistical confidence metrics. Expected outcomes include a fully operational low-cost pothole mapping workflow, an annotated dataset for AI training, a geospatial database of road defects, and a scalable methodology suitable for municipalities and academic institutions. By demonstrating that reliable pothole detection and quantification can be achieved with accessible equipment, the study offers a practical contribution toward cost-effective, data-driven road maintenance and safer urban mobility. (AU)