Journal of Korean Society of Transportation ISSN:1229-1366(Print) 2234-4217(Online) 대한교통학회지

In recent years, the total budget of the central government has been increasing at a rate of 4.9% per year, but the maintenance budget of the road sector has been decreasing annually. In addition, with the revision of the Road Act and the establishment of the Framework Act on Sustainable Infrastructure Management, all local governments shall establish a five-year master plan to improve maintenance and performance. However, it is difficult for local governments with relatively small budgets to introduce expensive equipment for road management. In response, the purpose of this study is to present a road crack detection method using mobile mapping system (MMS) and drones that are relatively inexpensive with the use of artificial intelligence. High-precision road surface photographic data could be acquired using MMS and drones. Pre-processing was carried out to train the artificial intelligence depending on the characteristics of the data. Labeling was conducted according to the type of road cracks and was labeled based on where cracks were present, where none were present, or where roads were not. The training was done by modifying the basic convolution network model and the mobilenetv2 neural network. As a result of the training, the data classification accuracy using the mobilenetv2 algorithm was able to achieve 95% on MMS data and 78% on drone data. It can be hypothesized that a more high-performance crack classification accuracy might be achieved by utilizing MMS and drone equipment. In the future, road cracks will be systematically managed by local governments and facility management agencies based on algorithms developed in this study. It is also expected to benefit citizens who use user-centered software to use roads.