Application of Automatic Image Recognition in Pavement Distress for Improving Pavement Inspection
AbstractHigh-frequency road inspection is the key to maintain all levels of road quality and to avoid casualties caused by bad road in Taiwan, inspections depend on open contract manufacturers, and they provide high-frequency road inspections and inspections of ancillary facilities, the inspection frequency is from one day to one week, according to the requirements of different agencies. However, the inspection equipment of the manufacturers and the inspection data lacks follow-up applications or numerical conversions, such as PCI, and cannot be applied to big data to enhance the long-term conservation of roads. This study uses existing road inspection methods and existing equipment to develop a back-end image recognition inspection software, hoping to improve the inspection efficiency by adding automatically identify damage, import it into the PCI values of ASTM D6433-16, and export the road quality in numeric. This study uses a traffic recorder and an imaging device as the main hardware, using the relationship between the image from the film and the speed of the car to obtain the image of the complete road, using SLIC Superpixels algorithm, which with two stages of image grouping, the pavement damage in the image will be selected. Using damage classification to define patches, potholes, longitudinal cracking, and crocodile cracking, then import into PCI for calculation. The results of this research has good conformity characteristics with semi-automated pavement inspection software, by adjusting the relationship between image capture frequency and car speed, it can develop a comprehensive pavement inspection, although the effectiveness must be reduced due to the depth measurement limit of 2D images, and less meticulous than traditional manual inspections, it is closer to the manual detection than the current semi-automation, the labor and time costs can be reduced for PCI measurement. The future hopes to deepen the development of imaging devices and artificial intelligence learning, to enhance the effectiveness of this software.