Deep Learning–Based Automated Generation of Material Data with Object–Space Relationships for Scan to BIM
Sohyun Kim, Kwangbok Jeong, Taehoon Hong, Jaehong Lee, Jaewook Lee
Abstract
Conventional scan to building information modeling (BIM) automation mainly deals with geometry. However, one of its limitations is the time it takes and the costs in generating material. Therefore, this study proposes an automated scan-to-BIM method considering both the geometry and material of building objects. It recognizes the geometry from a point cloud and the material from panorama images through deep learning–based semantic segmentation. The two extracted pieces of data are merged, and the BIM objects with material are automatically generated by using Dynamo. Here, the object–space relationships were applied to increase the accuracy of the material data to be included in the BIM object. As the result, the accuracy was improved by 48.66% compared with before the application. The proposed method can contribute to the improvement of the as-built BIM model usability because it can automatically generate a BIM model by reflecting the material, as well as the geometry of the existing building.