PAVES: Pothole Acquisition with Vision, Elevation, and Severity

PAVES (Pothole Acquisition with Vision, Elevation, and Severity) is a specialized RGB-D dataset for pothole detection, geometric characterization, and severity classification in road infrastructure. The dataset was acquired using a vision and depth sensing system based on a Kinect v2 sensor and an embedded acquisition platform, under real roadway conditions. It includes RGB images, depth data, digital elevation representations, bounding box annotations, and a structured metadata file containing geometric descriptors such as maximum depth, equivalent diameter, average diameter, and severity level. The potholes are categorized into severity classes according to measurable geometric criteria, enabling its use for computer vision, machine learning, pavement assessment, and intelligent transportation research. This dataset is intended to support the development and evaluation of models for automated pothole detection and severity estimation.