Indoor structure segmentation is a key step in the scan-to-BIM, but current point cloud segmentation methods focus on point cloud geometry or semantic segmentation. Therefore, this study proposes a point cloud segmentation method specifically for indoor building structures, identifying and segmenting indoor structures from indoor point clouds without segmenting indoor objects, which reduces the waste of computational resources, improves the degree of automation, and the efficiency and accuracy of scan-to-BIM.
This study proposes a geometry-driven segmentation framework specifically designed for indoor structural point clouds. The method introduces a novel corner feature (CF) to capture critical geometric transitions, enabling robust segment of vertical structures such as walls. Instead of relying on object semantics, the approach leverages ceiling geometry and elevation cues to differentiate structural elements (walls, floors, ceilings) from non-structural clutter. The overall design emphasizes lightweight computation, spatial reasoning and generalizability across diverse indoor environments.
Experimental results on the Stanford Large-Scale 3D Indoor Spaces (S3DIS) and Matterport3D datasets confirm that ceiling boundary geometry, combined with elevation-based filtering, can robustly delineate architectural structures in cluttered indoor environments. The CF proves effective in detecting corner points and enabling the segmentation of both planar and curved vertical surfaces with high accuracy.
This study focuses on the structural segmentation step within the scan-to-BIM process and proposes a lightweight, generalizable method based on contour features (CF) and elevation data. The method simplifies the workflow of indoor structural segmentation, enhances accuracy, improves BIM generation automation and supports the digital transformation of indoor building practices.
