Vision-Based Automated Pavement Distress Inspection: A Review
Agus Mulyanto, Riri Fitri Sari, Abdul Muis, Ruki Harwahyu
Abstract
Computer vision and image processing have been extensively researched to enhance road inspection systems, particularly for detecting pavement distress and assessing road surface conditions. However, most research on vision-based pavement defect inspection is experimental, with few comprehensive studies aligning with standards like ASTM D6433-18. This paper provides a review of vision-based automated pavement distress inspection systems for asphalt concrete. It examines four main aspects, including image acquisition, pre-processing, pavement distress inpsection methods, and pavement distress dataset. The discussion covers image acquisition methods, which include platforms such as vehicles, drones, and satellites, as well as vision sensors, sensor calibration, and techniques for capturing road surface images. Pre-processing techniques are discussed for improving image quality and isolating regions of interest. Pavement distress inpsection methods employ 2D and 3D approaches to classify different types of pavement distress, severity level estimation, and distress quantification. Additionally, the review provides an overview of existing pavement distress datasets, covering image collection, annotation, augmentation, and evaluation processes. This paper offers insights into the current state of research and references for designing vision-based automated pavement distress inspection systems that meet ASTM D6433-18 standards.