An Automated Assessment Technique for Pavement Defects Using a Laser Scanner and Deep Machine Learning

Al-Mistarehi, Bara'; Shtayat, Amir; Imam, Rana; Abdallah, Ashraf · 2025 · Crossref

DOI: 10.28991/cej-2025-011-03-015

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the limitations of traditional pavement inspection methods, which are often time-consuming, labor-intensive, and hazardous. The authors propose an automated assessment technique that integrates terrestrial laser scanning with deep machine learning to accurately identify, classify, and quantify pavement defects, specifically longitudinal cracks and rutting. The research aims to provide a rapid, precise, and safe alternative for monitoring road conditions, using data collected from highways in New Aswan City, Egypt. The methodology involved scanning two specific road sections using a FARO FOCUS S120 terrestrial 3D laser scanner. This device captured high-density point cloud data, recording the three-dimensional coordinates (x, y, z) and color intensity of the pavement surface. For longitudinal cracks, over 1.2 million points were scanned, while the rutting section yielded more than 4.5 million points. The data was processed using FARO SCENE and CloudCompare software to register scans and extract oriented data. To validate the automated findings, the authors conducted manual inspections using the Pavement Condition Index (PCI) procedure via the PAVER system. This involved visually assessing distress types, severity levels (low, medium, high), and quantities across randomly selected sample units. The laser-derived data served as input for five distinct machine learning models: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT). These models were trained to classify pavement conditions into three categories: normal, longitudinal cracks, and rutting, using a 70% training and 30% testing data split. The results demonstrated that the integrated approach effectively detected and classified pavement defects with high accuracy. The LSTM model achieved the highest performance with approximately 93% accuracy, followed by GRU at 91%, RF at 85%, SVM at 84%, and DT at 82%. The laser scanner successfully quantified the geometric characteristics of the defects, such as coordinate changes and depth, which correlated with the manually assessed PCI ratings. The manual PCI evaluations confirmed the presence of various distresses, including alligator cracking, potholes, and bleeding, with severity levels ranging from low to high, providing a ground-truth baseline for the machine learning predictions. The significance of this research lies in its demonstration that combining 3D laser scanning with deep learning offers a superior alternative to conventional visual inspections. The high accuracy rates, particularly of the LSTM and GRU models, suggest that this automated technique can reliably detect and classify pavement distresses, reducing safety risks for inspectors and minimizing traffic disruption. By providing precise, quantitative data on defect severity and location, this method supports more efficient maintenance planning and resource allocation for road networks. The study validates the potential of integrating advanced sensing technologies with artificial intelligence to enhance the precision and speed of infrastructure monitoring.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction