Point Cloud Segmentation Method Considering Wall Finishing Information Using 2D Material Segmentation and Back Projection

초록

Progress monitoring and quality control using as-built Building Information Modeling (BIM) are actively applied to construction industry. In order to effectively perform these management works, Scan-to-BIM is a key process to create as-built BIM models. In the Scan-to-BIM process point cloud segmentation is a critical task to identify object semantic information from point cloud data. While segmentation methods of main structural components such as walls, slabs, columns, and ceilings are actively studied and used for the management works, segmentation considering the finishing works of these components is still challenging. Therefore, this study proposed a point cloud segmentation method that considered wall finishing information, utilizing both point clouds and 2D images acquired from terrestrial laser scanners. The proposed method is composed of three main steps: 1) Segmenting as-built point clouds of main structural components through the comparison with as-planned BIM. 2) Applying a SegFormer material segmentation model that trained with wall finishing data (2D images) from terrestrial laser scanners to segment wall finishing information in 2D images. 3) Labelling the point cloud with recognized wall finishing information using back projection based on camera pose data. The proposed method is expected to contribute to the enchantment of the level of details (LoD) in as-built BIM and be useful in progress monitoring and quality control of finishing works. © 2024, Korea Institute of Construction Engineering and Management. All rights reserved.

키워드