Automated pavement distress detection for network-level management

 

Saved in:
Bibliographic Details
Author: Montúfar Chiriboga, Gabriel Jesús
Status:Versión publicada
Publication Date:2026
Description:Automated pavement distress detection has evolved from threshold- and texture-based image processing to deep networks for classification, object detection, and semantic segmentation. Nevertheless, high performance on closed datasets does not ensure that model outputs can support network-level maintenance decisions. This systematic review examined evidence from vehicle-mounted cameras, smartphones, three-dimensional sensors, and unmanned aerial vehicles, with emphasis on geographic generalization, severity quantification, and integration with pavement management systems. Specialized publisher databases and indexes were consulted for 2011-2026, and 40 primary studies were selected through explicit eligibility, extraction, and quality criteria. Object detectors were found suitable for rapid inventories of cracks, potholes, patches, and related defects, whereas segmentation and three-dimensional data were more appropriate for estimating extent, width, area, depth, or volume. Smartphones provide the strongest cost-to-coverage ratio for frequent screening; unmanned aerial vehicles add value for targeted surveys, inaccessible locations, and post-event inspection; and three-dimensional systems remain necessary when geometry drives the maintenance decision. Four persistent limitations were identified: domain shift across countries and seasons, incompatible distress taxonomies, limited external validation, and weak traceability from predictions to condition indices and maintenance treatments. A tiered architecture is proposed that combines mobile screening, selective reinspection, human audit, and georeferenced updating of the management system. The operational objective should not be the maximization of a laboratory metric, but the production of repeatable, calibrated, and economically useful observations per kilometer of road network.
Country:Portal de Revistas UCR
Institution:Universidad de Costa Rica
Repositorio:Portal de Revistas UCR
Language:Español
OAI Identifier:oai:portal.revistas.ucr.ac.cr:article/11929
Online Access:https://revistas.ucr.ac.cr/index.php/rvial/article/view/11929
Keyword:aprendizaje profundo
deterioro de pavimentos
gestión de pavimentos
imágenes
teléfonos móviles
vehículos aéreos no tripulados
deep learning
images
pavement distress
pavement management
smartphones
unmanned aerial vehicles