Image-Based Pavement Type Classification with Convolutional Neural Networks

Road pavement type classification is important in route planning, road maintenance and for autonomous vehicles. In this paper, we propose a deep learning based method for automatic road type classification from road surface images. The resulting binary classifiers (paved and non-paved road classes) achieve up to 98% classification accuracy on the test set that contains over 100 000 real-world road images that cover a distance of over 300 km.