Abstract
In the road surface inspection, the presence or absence of cracks is judged by visual interpretation of the road surface image taken. This process occupies about 20% of the whole process, which is a bottleneck in production. In this research, we developed a classifier that classifies cracked images using machine learning techniques, and improved efficiency by replacing visual interpretation with machine interpretation. The road surface image of the road managed by municipalities was divided into 50 cm squares, learning was carried out with about 140 thousand images, and the evaluation was carried out with about 15 thousand images, and the correct answer rate was 90.6% . Next, using this same classifier, judgment on the presence or absence of cracks was made on about 6 million images of prefectural control roads, and crack rate was calculated by aggregating every 20m and 100 m. The correlation coefficient between the value obtained by visual interpretation and the value obtained by machine interpretation was as high as 0.875. The time spent on machine interpretation was reduced to about 40% of the time required for visual interpretation. In addition, the actual working time of human beings has been reduced to about 2%.
