Synthetic Floor Plan Symbols for Accurate BIM Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems fail to efficiently convert graphical documents, particularly PDF files, into Building Information Model (BIM) elements due to lack of semantic information and data scarcity, requiring manual drafting and re-creation of electrical symbols.

Innovation Solution

A deep learning-based pipeline that automatically parses design drawings to extract geometric and semantic information of symbols, generating synthetic datasets to address data scarcity and privacy issues, and converts drawings to BIM elements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual drafting and re-creation of electrical symbols is used, then semantic information can be accurately captured, but productivity is reduced and time is lost

Engineering Contradiction:

Improveaccuracy of semantic informationVSAvoiddrafting efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent uses PDF files containing vectorized information of electrical symbols as templates or copies. The system automatically extracts and reuses this existing visual information rather than requiring manual re-drawing, thereby maintaining accuracy while improving productivity.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent replaces the manual mechanical drafting process with an automated computer-based system using deep learning and image processing algorithms. This substitution eliminates manual labor while preserving the ability to capture semantic information through automated symbol recognition and classification.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If automated conversion systems are used, then productivity is improved, but manufacturing precision deteriorates due to lack of semantic information

Engineering Contradiction:

Improveconversion speedVSAvoidaccuracy of BIM element extraction

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent transforms the input data by converting PDF files into multiple image formats (JPEG, PNG, BMP) and applying various image processing parameters such as grayscale conversion, thresholding, and noise filtering. These parameter changes enhance the quality of input data for the deep learning model, thereby improving extraction accuracy while maintaining automated processing speed.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent performs preliminary processing of PDF files including conversion to images, grayscale transformation, and noise removal before the main extraction process. This preliminary action prepares the data in optimal form for the deep learning model, improving subsequent extraction precision without compromising overall productivity.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If more training data is collected, then machine learning model precision is improved, but loss of time increases due to data collection and manual labeling

Engineering Contradiction:

Improvemodel prediction accuracyVSAvoiddata preparation time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent generates synthetic training data by copying and transforming existing floor plan images through various geometric transformations (rotation, scaling, flipping) and adding synthetic labels automatically. This approach creates large amounts of training data without requiring manual labeling, thereby improving model accuracy while minimizing time investment.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system performs self-labeling of synthetic training data through automated rule-based assignment of labels based on geometric properties and spatial relationships in the transformed images. This self-service approach eliminates the need for manual annotation, allowing the system to generate its own training data independently and efficiently.

Inventive Principle:

Principle #25Self-service

4Quantity of substance

If synthetic data generation is implemented, then data scarcity is addressed, but device complexity increases

Engineering Contradiction:

Improveamount of training dataVSAvoidsystem architecture complexity

Core Design Contradiction:

Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent divides the synthetic data generation process into distinct modular components: image transformation module, label generation module, and data augmentation module. Each module handles a specific aspect of synthetic data creation, making the overall complex system manageable and maintainable while still producing large volumes of training data.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12518066B2Synthetic data generation for machine learning tasks on floor plan drawings

Publication Date: 2026.01.06 AUTODESK INC

AI summary

A method and system provide the ability to generate and use synthetic data to extract elements from a floor plan drawing. A room layout is generated. Room descriptions are used to generate and place synthetic instances of symbol elements in each room. A floor plan drawing is obtained and pre-processed to determine a drawing area. Based on the synthetic data symbols in the floor plan drawing are detected. Orientations of the detected symbols are also detected. Based on the detected symbols and orientations, building information model (BIM) elements are fetched and placed in the floor plan drawing.