REFLECTION OF THE POSITION OF CRACKS IN A 3-D MODEL OF A BRIDGE USING SEMANTIC SEGMENTATION

抄録

When constructing a 3-D model of a bridge for maintenance purposes, it is considered easier to confirm the position of the damage and the shape of the model by reflecting the actual location of the damage in the model. Cracks are typical damage, and there have been many studies on methods of automatically detecting cracks from images of the concrete surface. However, it is still difficult to detect the cracks if nonconcrete objects are reflected in the image. This study makes it possible to detect cracks from a number of captured images by extracting in advance the concrete area only, using semantic segmentation from the images of the bridge. Furthermore, based on the image captured in detecting a crack, it was possible to construct a 3-D model of the bridge using Structure from Motion, and to construct a 3-D model in which the positions of the cracks were reflected. This paper is the English translation of the authors’ previous work [Yamane, T. et al. (2020). “Reflection of crack locations in 3-D models of bridges using semantic segmentation” Intelligence, Informatics and Infrastructure, Vol. 1, No. J1, 491-497 (in Japanese)].