Automatic Classification and Quantification of Basic Distresses on Urban Flexible Pavement through Convolutional Neural Networks

Abstract

[EN] Pavement condition assessment is a critical step in road pavement management. In contrast to the automatic and objective methods used for rural roads, the most commonly used method in urban areas is the development of visual surveys usually filled out by technicians that leads to a subjective pavement assessment. While most previous studies on automatic identification of distresses focused on crack detection, this research aims not only to cover the identification and classification of multiple urban flexible pavement distresses (longitudinal and transverse cracking, alligator cracking, raveling, potholes, and patching), but also to quantify them through the application of Convolutional Neural Networks. Additionally, this study also proposes a methodology for an automatic pavement assessment considering the different stages developed in this research. This methodology allows for a more efficient and reliable pavement assessment, minimizing the cost and time required by the current visual surveys.