DEVELOPMENT OF PAVEMENT INSPECTION SYSTEM BY USING CAR-MOUNTED CAMERA AND U-NET MODEL

Abstract

 We have developed a pavement inspection system by using car-mounted camera and U-net model which is one of the deep learning. First, the cracks on this image were sketched by hand and the teacher data for model learning was constructed. By applying the model to the road surface image, cracks can be detected with high accuracy and speed. Next, in order to verify the accuracy of the crack evaluation using this system, we used the correct value data from the road surface property measurement vehicle. As a result of analysis from a total of 500 blocks of sample data, it was shown that the detection rate (Level II+) was rank A (80-100%), the detection rate (Level III), the hit rate (Level II+), and the hit rate (Level III) were rank B (60-80%) which proved to have sufficient inspection accuracy.