Floorplan Extraction via Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for extracting information from architectural floorplans are inefficient and resource-intensive, particularly when dealing with large sets of stored documents, as they often require manual browsing and are limited by the need for specific rules or training data for each type of feature.
Innovation Solution
The use of multiple smaller machine learning models to extract different types of semantic information from floorplans, including boundary segmentation and room segmentation, allows for efficient identification and extraction of floorplans from large datasets, reducing the overhead of training and execution compared to traditional approaches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual browsing and rule-based extraction methods are used to extract floorplan information from architectural drawings, then the extraction process can be performed with simple tools, but the time consumption and resource overhead increase significantly when dealing with large document sets
Solution Approach 1:
The patent replaces manual browsing and rule-based extraction (mechanical systems) with machine learning models that automatically extract floorplan elements. The system uses trained neural networks to identify walls, doors, windows, and rooms directly from image inputs, eliminating the need for manual document browsing and rule-based processing, thereby significantly reducing time consumption while maintaining ease of use through automated pipelines.
Solution Approach 2:
The machine learning models are trained once on diverse floorplan data and then autonomously perform extraction tasks without requiring manual intervention or document-by-document rule configuration. The system serves itself by automatically adapting to different floorplan styles and conventions through the generalized knowledge embedded in the trained models, reducing ongoing resource overhead.
2Device complexity
If a fixed set of rules or single machine learning model is used for feature extraction, then the system structure remains simple, but the system cannot generalize to identify different types of features beyond the training data
Solution Approach 1:
The patent segments the feature extraction task into multiple specialized machine learning models, each trained to identify specific floorplan elements (walls, doors, windows, rooms). This segmentation allows each model to focus on particular features while collectively providing comprehensive extraction capabilities. The modular architecture maintains manageable system complexity while significantly improving adaptability to different feature types through ensemble prediction.
Solution Approach 2:
The machine learning models are designed with universal architectures that can be trained on diverse datasets and applied to various floorplan styles and conventions. The systems use transfer learning and are trained on comprehensive datasets that encompass multiple architectural styles, enabling the models to generalize to feature types and conventions not explicitly present in the training data, thereby achieving multi-functionality across different extraction scenarios.
3Adaptability or versatility
If multiple machine learning models are used to extract different semantic information, then the variety of extractable features increases, but the training and execution overhead increases compared to single-model approaches
Solution Approach 1:
The patent divides the extraction task into specialized models for different feature types (walls, doors, windows, rooms), allowing each model to achieve high accuracy for its specific function. This segmentation enables parallel execution of models, which can reduce overall execution time compared to a single sequential model. The modular structure also allows for efficient resource allocation and selective model execution based on document type.
Solution Approach 2:
The patent merges the outputs of multiple specialized models into a unified extraction result through ensemble prediction. The system combines predictions from wall detection, door detection, window detection, and room segmentation models to produce comprehensive floorplan annotations. This merging approach leverages the strengths of each specialized model while maintaining a coordinated system that manages training and execution overhead through shared preprocessing and postprocessing infrastructure.
Data Source
AI summary
One embodiment of the present invention sets forth a technique for extracting data from an architectural drawing. The technique includes performing one or more operations via one or more machine learning models to extract a first image of a floorplan area from the architectural drawing. The technique also includes generating a boundary segmentation based on the first image of the floorplan area, wherein the boundary segmentation includes one or more boundary types for one or more portions of the floorplan area.