Automated Distress Detection and Measurement in Urban Asphalt Pavements Using Deep Learning

Road pavement condition is crucial information for maintaining infrastructure integrity, assuring road safety, and optimizing maintenance costs. However, historical condition data of urban pavements condition are not usually available because there is no technology that can evaluate pavement conditions in a low-cost and efficient way. On the one hand, this research proposes a system capable of obtaining and processing pavement images to evaluate urban pavements. On the other hand, a deep learning model is trained with over 50,000 images of 13.2 m × 2.6 m of asphalt pavement from different zones of Santiago, Chile. Following the processing of these images, the following distresses were manually labeled with two different levels of severities: patches; potholes; and transversal, longitudinal, and fatigue cracking. Finally, the distresses are measured using the dimensions of the artificial neural network’s bounding boxes. The artificial neural networks (ANNs) proposed for this research are YOLOv5 and YOLOv7.