Automated Pavement Condition Assessment using Unmanned Aerial Vehicles (UAVs) and Convolutional Neural Network (CNN)

Abstract

Assessing pavement condition is extremely essential in any effort to reduce future economic losses and improve the structural reliability and resilience. Data resulting from pavement condition assessment are used as a record of infrastructure performance and as a major component to assess their functionality and reliability. However, pavement condition assessment is challenging because of the cost associated with assessment, safety issues, and the accessibility restrictions, especially after natural hazards. This research aims to develop an automated classification model to rapidly classify pavement distresses. High-resolution aerial images representing alligator and longitudinal cracks are collected for flexible pavements using Unmanned Aerial Vehicle (UAV) around East Carolina University (ECU) campus. The image classification model is developed using Convolutional Neural Network (CNN), a deep learning approach. The results of the developed model indicate an accuracy of 96.7% in classifying the two categories of pavement distress. The developed model was further tested on a set of test images yielding a prediction accuracy of 90%. The methodology behind the developed model will help to reduce the need for on-site presence, increase safety, and assist emergency response managers in deciding the safest route to take after hurricane events. Additionally, application of the model will enable transportation engineers in rapidly assessing the pavement damage, aid in making quick decisions for road rehabilitation and recovery and devise a restoration or repair plan.