Automatic Indoor Building 3D Wireframe Modeling from Point Clouds Spatial Geometry Information for As-Built BIM

The complexity of indoor structures and various limitations of point clouds make automatic indoor three-dimensional (3D) wireframe modeling for 3D as-built building information modeling (BIM) suffer multiple challenges. This paper presents a novel indoor building 3D wireframe modeling method from point clouds spatial geometry information for as-built BIM. The main innovation is to improve the semantic labeling accuracy of small-sample structural elements (SSSEs) and the line segment extraction accuracy of small plane and detail structures. A semantic labeling network considering point contextual features was first used to classify indoor point clouds as different types of structural elements. Then boundary points of different structures were extracted using dimensional features and spatial geometric relations. To obtain a regular and complete 3D wireframe model, an optimization method based on spatial relative position was performed. Finally, the 3D wireframe model was imported into the modeling tool for as-built BIM modeling. The 3D wireframe models of typical indoor scenes were evaluated and compared with different methods. The analysis of important modules and parameters was verified in the ablation experiments, which further validated the effectiveness and robustness of our method. Experiments show that our method significantly improves the labeling accuracy of SSSE (beams, columns, doors, and windows) and constructs 3D wireframe models with higher fitting accuracy and more detailed structural information.