Development of a road defect inspection system (AI–RDIS) using AI deep learning technology

Abstract In the past, road contractors primarily used vehicle-mounted equipment for routine road defect inspections, and inspectors logged and communicated their findings in writing or over the phone. Over time, the shortening of inspection periods, expansion of inspection areas, and diversity of inspection operations fueled the advancement of technologies used for road defect inspections. In recent years, emerging artificial intelligence (AI) technologies have been adopted to assist in this work. This paper introduces the ongoing efforts in Taipei City, Taiwan, since 2016 to collect road defect images using vehicle-mounted equipment. The city’s road contractor incorporated the You Only Look Once (YOLO) algorithm, which utilizes deep learning for road defect image training, and launched the AI Road Defect Inspection System (AI–RDIS). The system was tested to evaluate the applied deep learning algorithms and the performance of defect recognition. Currently, Taipei City employs the AI–RDIS to automatically recognize road defects in daily road defect inspections.