Detection and classification of cracks on road pavement in highway by deep convolutional neural networks

Abstract

Road Eye is a special vehicle for inspecting various road conditions under driving and it is mainly employed in highway at the east area in Japan. One of the measured quantities is a continuous road surface image captured by the line cameras and health of the road surface is manually evaluated from state of cracks in the image according to several complex rules. The final objective of the study is to automatically evaluate health of the road surface by applying several deep convolutional neural networks to the multi-scale images, and as the first step of the study, this paper presents two neural networks for semantic segmentation. The first one is a network for distinguishing road surface, expansion-joint and grooving in the image in wide area. Then, from the road surface in local area, the second one detects and classifies cracks according to its depth. Both networks show accuracies equivalent to the manual judgements.