Abstract
Indoor point cloud data is easy to obtain and contains a large amount of data, but there is no connectivity, no structural attributes, and no semantic information between point clouds. It is very difficult to extract interior components. In order to solve the problem that it is difficult to extract the closed door in the indoor point cloud, and the indoor scanning point cloud data is noisy and redundant, a point cloud segmentation method is proposed to reduce the amount of data and the impact of noise. Because most of the indoor components are not on the same horizontal plane, different component information can be obtained by slicing the point cloud at different heights. In these slices, the component that always exists is the wall. By comparing other data with the point cloud data with a more complete wall, the required component can be extracted. The experimental results show that this method can accurately extract the indoor door component information and beam component information, providing a data basis for building model refinement.
Introduction
While 3D laser scanning makes it easy to capture the geometry of a room, the resulting "point cloud" is just a massive collection of coordinates. It lacks semantic meaning; the computer doesn't inherently know which points belong to a wall and which belong to a closed door or a structural beam. Researchers have developed a new method called Hierarchical Slicing and Stacking Analysis (HS-OA) to automatically extract these architectural features. By "slicing" the building at specific heights and subtracting the common wall structures, the system can isolate components that are otherwise nearly impossible to distinguish from flat surfaces—even when doors are completely closed.
Existing extraction methods often struggle with noise from furniture, the difficulty of identifying closed doors (which look like walls), and the neglect of structural beams. This research addresses these gaps by leveraging the vertical distribution of architectural elements.
Methodology
The authors propose a multi-step workflow to "clean" and then "dissect" the indoor space.
Floor Division and Noise Reduction
The system first projects all points onto the Z-axis (height). Since floors and ceilings are flat, they create massive "peaks" in the density histogram, allowing the software to automatically separate a multi-story building into individual floors. To handle indoor clutter, the researchers observed that most "noise" (desks, chairs, etc.) exists below 1.2 meters. By focusing on slices above this height, they significantly improve data quality.
Hierarchical Slicing and Overlay Analysis (HS-OA)
The core innovation involves taking horizontal slices at strategic heights. For example, a slice at 2.0m–2.5m captures the upper frames of doors, while a slice at 3.6m captures ceiling beams.
HS-OA is essentially "point cloud subtraction." The system compares a slice containing only walls with a slice containing the target component (like a door top). Using a K-neighborhood search, it identifies points in the "component slice" that don't exist in the "wall slice" and preserves them as the extracted feature. Because walls exist at almost every height, they serve as a perfect baseline.
Experimental Setup
The study utilized a dataset captured using a Trimble TX5 ground laser scanner at Building F of the Beijing University of Civil Engineering and Architecture (BUCEA). The scale included two floors with 14 rooms and connecting stairwells, totaling approximately 909MB of raw point cloud data.
Key Results and Insights
The experimental results demonstrated high reliability for indoor modeling:
- Door Extraction: The method achieved 100% accuracy in the test environment, successfully identifying all 27 doors across two floors. By focusing on the upper edge of the door frame, the method identifies doors regardless of whether they are open or closed.
- Beam Extraction: The system accurately identified the number of crossbeams in five out of six tested rooms. One beam was missed due to a small plane width falling below the detection threshold.
- Architectural Standards: The method succeeds by incorporating human knowledge about building standards—specifically that doors and beams appear at predictable vertical intervals.
Limitations and Future Work
Despite the high accuracy, the authors noted a few hurdles. Complex stairwells proved too difficult for automated slicing and required manual removal. Additionally, other ceiling-mounted objects, such as hot water pipes, can be mistaken for beams. Future work will focus on more robust semantic filtering to distinguish structural beams from utility fixtures.
Summary updated