Road Traversing Knowledge for Quality Classification

This dataset consists of 6.297 labeled images from Brazil's inner country roads captured by a low-cost camera (HP Webcam HD-4110) attached to a moving vehicle. The dataset presents three quality levels classes (good, regular, and bad) for each surface kind: asphalt, paved and unpaved. The only exception is the unpaved surface that does not contain the class "good" due to its poor nature. The number of images by surface-quality class is:

asphalt-good: 1978
asphalt-regular: 839
asphalt-bad: 464
paved-good: 1179
paved-regular: 324
paved-bad: 124
unpaved_bad: 593
unpaved-regular: 796

All images were collected during the daytime in good weather conditions. The frames present lighting variations, shadows, and solar glare. Moreover, it is also noticed intra-class variations, such as asphalt surfaces with different color intensities, e.g., older asphalts are usually lighter than the new ones; paved surfaces formed by different patterns, e.g., cobblestone and concrete; and unpaved surfaces, which are naturally uneven.

The directory is organized by subdirectories named with the "surface_quality".