PaveDistress: A comprehensive dataset of pavement distresses detection.

Abstract

The PaveDistress dataset contains high-resolution images of road surface distresses, including cracks, repairs, potholes, and background images without defects. The data were collected using a specialized pavement inspection vehicle along the S315 highway in China. The vehicle was equipped with a Basler raL2048-80km line scan camera and infrared laser-assisted lighting, capturing images at 1mm intervals with a resolution of 3854 × 2065 pixels. The images were taken every 2 meters across various lighting conditions, including daylight, dusk, and in challenging environments such as tunnels and cloudy weather. The dataset is organized into distinct categories, covering transverse cracks, longitudinal cracks, map cracks, and more, enabling detailed categorization of pavement distresses. Each image represents a real-world road coverage area of 3.9m × 2.1m, allowing for accurate measurements of defect dimensions. This dataset supports the development of deep learning models for non-destructive detection of road defects, providing valuable resources for civil engineering research and practical applications in road maintenance systems. The dataset can be reused for tasks such as image classification, object detection, and segmentation, enabling researchers to create advanced machine learning models for road distress detection and assessment. By providing high-quality, diverse images, the PaveDistress dataset offers significant potential for research in automated pavement condition monitoring and management systems.