Automated Detection and Estimation of Pavement Distress Dimensions Using Computer Vision

Highways play a crucial role in connecting cities. Pavement issues can have a direct impact on the economy, leading to costs related to accidents, cargo transportation delays, compensation for victims, increased product prices, and various other factors. The assessment of pavement surface distress is a fundamental step in pavement management to determine its degree of deterioration. Traditionally, many companies and government entities conduct this assessment manually, where assessors record the characteristics of distresses on foot or in slow-moving vehicles. This manual method is impractical, which is why our focus is on automated methods, so cameras are mounted on vehicles to record pavement conditions through videos or photographs. In this article, we have developed a deep learning computational model to detect distresses in pavement surface using a dataset collected by the authors. This model achieved an 84% accuracy in classifying distresses after neural network training. Additionally, we propose another model to estimate the dimensions of detected distresses. This study aims to present a systematic approach to the automated distress automated process on highways, considering distress types, frequency, and severity. It can serve as a decision support tool for researchers, engineers, analysts, and pavement management professionals for future assessments.