AI-Powered Floor Plan Reading

Introduction

Floor plans are critical tools in architecture, engineering, and real estate industries, providing visual representations of building layouts. Historically, these diagrams were manually analyzed by experts. However, with advancements in computer vision (CV) and artificial intelligence (AI), automated extraction and interpretation of data from floor plans have become feasible. This case study focuses on leveraging computer vision techniques to automate the reading and analysis of floor plan images.

Problem Statement

Real estate professionals often struggle to analyze large volumes of floor plan images efficiently. Manually extracting relevant information from these plans, such as room dimensions, layouts, and features, is time-consuming, prone to errors and huge cost.

The Solution: AI Powered Floor Plan Reading

OpenCV (Open Computer Vision) is a powerful open-source library for computer vision tasks. Combined with AI techniques, it can be effectively used for floor plan reading.

Key Steps for Floor Plan Reading:

• Image Loading and Preprocessing:
• OpenCV is the best choice for reading images using the imread function. Convert the BGR colored image to GRAY color image using the cvtColor function. For Blurring image medianBlur function is used. Then finally adaptiveThreshold function is used with “ADAPTIVE_THRESH_GAUSSIAN_C” and “THRESH_BINARY_INV” for preprocessing the floor plan real image.
• Contour Detection:

• For finding contours from the Floor Plan image, used findContours function of openCV with “RETR_LIST” and “CHAIN_APPROX_SIMPLE” parameters. Where chosen minimum area of contour > 100, for dropping unnecessary contours.
• Detected parts of Floor Plan Image and it’s area:

• The most important and critical part of the floor plan is detecting multi-corner point rooms. Where each room area > 10000. For window detection its area > 2000 and area < 8000 and the detected point must be 4. Finally for door detection, where detection contour point is 3 and area < 8000.
• Room data process:

• For the multi-corner point of the room, need x,y value for each corner, find the magnitude value of x,y then square the magnitude. Find the normalized value from magnitude. Finally, normalizedPositions and direction of the room data.
• Window data process:

• For 4 point corner/contour points, need x,y value of each corner point. Find the center x,y point from each 4 corner point. Calculate magnitude from center point. And then square magnitude from the magnitude. Calculate normalized values from center point and magnitude. The most important part is finding the direction vector for the window's front part.
• Door data process:

• Like the previous window data process, find x,y coordinates points, Calculate magnitude and square magnitude from coordinates values. Finally finding normalized positions and direction vectors.
• API build and prediction:

• We used Django for building API. Which will take one image input at a time and provide a JSON file that is generated based on an imputed floor plan image. This JSON file will push to the Unity 3D environment. Finally, we can see the 3D floor plan in the Unity 3d environment.

Benefits of AI-Powered Floor Plan Reading

AI-powered floor plan reading using OpenCV Python offers several advantages and benefits:

1. Automation and Efficiency:

• Streamlines processes: Automates the tedious task of manually analyzing floor plans, saving time and effort.
• Reduces human error: Minimizes mistakes that can occur during manual interpretation.

2. Accuracy and Precision:

• Improved data extraction: Accurately extracts relevant information such as room dimensions, layouts, and features.
• Enhanced decision-making: Provides precise data for informed planning, design, and construction.

3. Scalability and Flexibility:

• Handles large volumes: Can process numerous floor plans efficiently, even with complex layouts.
• Adapts to various formats: Accommodates different floor plan types and styles.

4. Cost-Effectiveness:

• Reduces manual labor: Eliminates the need for extensive manual labor, leading to cost savings.
• Improves efficiency: Increases productivity and streamlines workflows.

5. Data-Driven Insights:

• Generates valuable data: Extracts insights from floor plans, enabling data-driven analysis and decision-making.
• Supports informed planning: Provides valuable information for urban planning, facility management, and real estate analysis.

6. Integration with Other Systems:

• Interoperability: Can be integrated with other software systems for seamless workflow automation.
• Enhanced functionality: Combines with other AI tools for more comprehensive analysis and applications.

Example use cases:

• Architecture and construction: Analyzing floor plans for design, construction, and space optimization.
• Real estate: Assessing property value, identifying potential renovations, and creating virtual tours.
• Interior design: Planning layouts, selecting furniture, and visualizing design concepts.
• Facility management: Optimizing space utilization, managing maintenance, and planning renovations.

The Challenges and Limitations

OpenCV, a powerful computer vision library, offers robust tools for processing and analyzing images. However, reading floor plans presents unique challenges due to their specific characteristics. Here are some key limitations:

1. Image Quality and Complexity:

• Blurry or Low-Resolution Images: Floor plans may be of poor quality, making it difficult for OpenCV to accurately detect edges and features.
• Complex Layouts: Intricate floor plans with numerous rooms, hallways, and details can overwhelm the algorithm, leading to inaccurate segmentation.

2. Geometric Distortion and Perspective:

• Non-Orthogonal Views: Floor plans are often not perfectly orthogonal, introducing geometric distortions that can affect feature extraction and measurement accuracy.
• Perspective Effects: If the floor plan is captured at an angle, perspective distortion can further complicate analysis.

3. Symbol Recognition and Interpretation:

• Custom Symbols: Floor plans use various symbols (e.g., doors, windows, furniture) that may not be readily recognized by standard object detection models.
• Contextual Understanding: Interpreting the meaning of symbols and their relationships within the floor plan requires domain-specific knowledge.

4. Data Variability:

• Different Drawing Styles: Floor plans can vary in drawing style, scale, and level of detail, making it challenging to develop a single model that works well for all types.
• Noise and Artifacts: Scanned or digitized floor plans may contain noise, artifacts, or imperfections that can interfere with the analysis process.

6. Incomplete or Inconsistent Data:

• Missing Information: Floor plans may lack essential details, such as room dimensions or specific features.
• Inconsistent Labeling: Labels for rooms, features, or symbols may be inconsistent or ambiguous, making it difficult to extract accurate information.

Technologies Used

List of the technologies used for this project are below:

• OpenCV: This is the main library in python for image preprocessing and reading floor plan images.
• PIL: PIL is another image processing library in python. This library have some awesome features like OpenCV for image processing
• Scikit-Image: Scikit-image is an open-source image processing library for python programming language. It includes algorithms for segmentation, geometric transformations, color space manipulation, analysis, filtering, morphology, feature detection, and more.
• AWS: AWS sage-maker is used for deploying floor plan reading applications.

Conclusion

The generated floor plans serve as a valuable asset for optimizing device placement, providing personalized recommendations, and enhancing the overall user experience of smart home systems. The success of this case study highlights the potential of AI and OpenCV to revolutionize various industries, including architecture, interior design, and real estate. By leveraging advanced computer vision algorithms, the company successfully developed a system capable of accurately generating floor plans from images or sketches.