Cycling Street Surface Dataset

The Cycling Street Surface dataset contains images of various street and road surfaces captured with an instrumented bicycle, to enable machine learning models to classify and monitor infrastructure conditions. The data was captured using an OV2640 camera module mounted on a bicycle. The images were cropped to two different resolutions (96x96 and 48x192 pixels) to facilitate deployment on edge devices with limited computational resources. We annotated the images into four distinct classes based on the type of street surface: asphalt, paving stone, sett, and unpaved. The last class is a combination of diverse unpaved surfaces such as dirt, gravel, and grass.