ABSTRACT
In Japan, the number of social infrastructure facilities such as bridges and tunnels, which were constructed during the period of high economic growth and have passed 50 years, is increasing. Inspection is important for their management. In recent years, many researches on inspection automation have been carried out. In particular, research on crack detection using image classification, object detection, and semantic segmentation has made remarkable progress. Despite these advances, it is still difficult to accurately describe the 3D location of damage because current periodic inspection records are mainly in 2D format. In this research, we constructed a crack detection system by combining YOLO, an object detection technology, and instantNeRF, a 3D scene generation technology, and compared the free-view image by instantNeRF with the image without instantNeRF from the same angle. These approaches demonstrate the potential for storing inspection results in a 3D format, enabling a more comprehensive and intuitive visualization of structural damage.
