Continuous Visual Survey of Road Pavement Using Convolutional Neural Networks and Smartphone Technology: A Data-Driven Approach

In order to reduce costs and time associated with road quality assessment, a Pavement Management System (PMS) must embrace survey methods that expedite processes and minimize the demand for expensive procedures. With the widespread use of smartphones, numerous studies leverage these tools to simplify measurement processes. Additionally, the integration of computer vision (CV) with machine learning and artificial intelligence (AI) techniques has enabled accurate detection of pavement defects, providing foundation for road maintenance planning with less resources. This article develops and evaluates the accuracy of an alternative Continuous Visual Survey (CVS) process using a smartphone mounted on a vehicle's windshield, coupled with AI tools for detecting cracks, patches, and potholes on road pavements. An Android application was developed for smartphones to capture photos while simultaneously collecting time and GPS data. The collected data were split into two groups, with the first group used for training Convolutional Neural Network (CNN) models and the second for testing. The developed model showed an average precision of 0.72, recall of 0.50, and a mAP (Medium Average Precision) of 0.54 in detecting defects on the pavement. Indicating the potential effectiveness of AI in accurately computing road distresses.