Architectural Floor Plans from Aligned Point Cloud Landmarks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating architectural plans from scans face challenges in computational complexity, processing time, and potential inaccuracies due to large datasets and misalignments, requiring significant computational resources and user expertise.
Innovation Solution
A computer-implemented method that reduces data requirements by focusing on user-defined key landmark points, allowing on-site verification and correction of scans, and simplifies the annotation process through a dedicated user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If densely populated point cloud data is used to mark wall positions, then detailed information is captured, but computational demand and data processing complexity increase significantly
Solution Approach 1:
The patent extracts only the essential features from dense point cloud data - specifically corner points and wall lines - rather than processing the entire dense dataset. This extraction approach maintains the necessary detail information while dramatically reducing computational complexity by focusing only on the critical geometric features needed for architectural plan generation.
Solution Approach 2:
The system creates simplified 2D representations (copies) of the 3D point cloud data at scanner height, working with these simplified copies rather than the original complex 3D data. This copying approach preserves the essential spatial relationships needed for accuracy while reducing processing complexity.
2Measurement precision
If multiple photographs are taken and stitched together to create a 3D model using photogrammetry, then highly accurate models can be produced, but the process requires significant time and effort
Solution Approach 1:
The system performs preliminary processing by extracting corner points and wall lines directly from the point cloud data before attempting to combine multiple scans. This preliminary extraction of key features reduces the subsequent processing time when stitching multiple photographs together, as the essential geometric information is already identified and ready for integration.
Solution Approach 2:
The patent segments the complex photogrammetry process into distinct steps: first capturing multiple photographs, then extracting corner points and wall lines from each, and finally combining them. This segmentation allows for more efficient processing by focusing computational effort on the critical feature extraction rather than processing all pixel data simultaneously.
3Extent of automation
If machine learning algorithms are used to identify wall positions automatically, then the process is automated, but the solutions struggle with complex or irregularly shaped rooms and require substantial computational power
Solution Approach 1:
The system introduces an intermediary step between automated scanning and final plan generation - manual verification and adjustment of corner points and wall lines. This intermediary approach combines the efficiency of automation with human expertise in handling complex geometries, allowing the system to maintain high automation levels while adapting to irregularly shaped rooms that pure machine learning struggles with.
4Measurement precision
If small errors occur in scan alignment or identification of structural elements, then significant inaccuracies result in the final floor plan, but these errors are difficult to identify and correct
Solution Approach 1:
The patent implements feedback mechanisms where the system combines multiple scans and compares the extracted corner points and wall lines across different scans. This comparison provides feedback that helps identify misalignments and errors, allowing users to detect and correct inaccuracies by viewing how features from different scans relate to each other in the combined model.
Data Source
Figure 1
Figure 2
Figure 3A~3C
AI summary
A computer-implemented method for generating architectural floor plans from scans, the method comprising: (a) acquiring a point cloud (110) representing the shape of a room, generated by a room scanner (15) located at a specific location (PA, PB, PC) within the room; (b) displaying a two-dimensional (2D) cross-sectional view of the point cloud (110); (c) prompting a user to draw lines between corners of walls over the 2D view to create an annotated point cloud with user-defined corner points (212) and wall lines (211); (d) repeating steps (a), (b) and (c) for a plurality of different locations (PA, PB, PC) within the same room; (e) combining the plurality of annotated point clouds by aligning them based on the user-defined corner points (212) and wall lines (211); (f) averaging the position of user-defined corner points (212) and wall lines (211) in case of spatial misalignment; and (g) outputting the architectural floor plan comprising walls detected based on averaged positions of scan points in the vicinity of user-defined corner points (212) and wall lines (211).