Artificial intelligence applications in pavement infrastructure damage detection with automated three-dimensional imaging – A systematic review
Saleh Abu Dabous, Mohamed Ait Gacem, Waleed Zeiada, Khaled Hamad, Rami Al‐Ruzouq
Abstract
Pavement damage detection is vital in ensuring road-users safety and reducing economic losses related to maintenance and road incidents, thus making it a highly discussed research topic. Non-destructive pavement damage inspection methods have been rigorously explored, compared and integrated in the road health and safety practices. In particular, 3D imaging stands out due to its ability to extract the depth and geometric characteristics of the detected defects, thus facilitating performing accurate maintenance and corrective actions. During the past few years, with the rise of the highly accurate and robust artificial intelligence-driven analysis, researchers have increasingly adopted machine learning and deep learning models in pavement damage assessment. This paper systematically and exhaustively reviews the use of artificial intelligence-based 3D automated damage detection of pavement performed using laser scanners, stereo cameras/structure from motion and infrared sensors. It compiled 85 contributions published between 2011 to mid-2024. From which, the adopted artificial intelligence models, utilized 3D data collection hardware, utilized datasets, 3D data pre-processing techniques and application domains were extracted, compared and critically analyzed. The survey ultimately highlights the gaps and potential future research directions. Key findings highlight the need to explore more cost-effective and advanced 3D data collection methods, such as drones, in addition to enhancing 3D data preprocessing techniques to boost AI performance. Furthermore, the lack of comprehensive open 3D datasets is a significant gap that, if addressed, could help standardize research. Moreover, the research highlights the need for expanding the use of AI models to cover a wider range of pavement deformities. • Reviewed AI-based 3D imaging of pavement damages. • Discussed related AI models, their merits, limitations and implementation. • Discussed 3D data acquisition systems, datasets and data pre-processing. • Discussed application domains related to different pavement damages. • Highlighted research gaps and suggested potential future research directions.