Preservation of civil engineering facilities, such as tunnels and manholes, has become important since such facilities are aging. In conducting quantitative structural health assessments based on inspection results, it is necessary to carry out structural calculations, which require 3D data that have the correctly mapped locations of deteriorated areas. Since panoramic images require fewer shots due to the wide view angle and reduced data management costs, we chose 3D reconstruction from panoramic images, which is suitable for obtaining 3D data of infrastructure facilities. Indoor-space 3D reconstruction needs boundary detection, which estimates floor-wall, ceiling-wall and wall-wall boundaries. Although existing models can be applied to room spaces, it is difficult to apply them to the inside of civil engineering facilities that have many objects. We propose a boundary estimation model that uses 3D point clouds obtained from drawings with panoramic images. Improvement in boundary detection accuracy enables the precise mapping of the deteriorated area in 3D, which will lead to the automation of structural calculations.