3D Building Reconstruction from Floor Plans

A web application that converts 2D floor plan images into interactive 3D models using classical computer vision algorithms and C programming.

Features

• Upload floor plan images (PNG, JPG, PGM, PPM)
• Automatic wall, room, door, and window detection
• Generate 3D models with and without roofs
• Interactive 3D visualization with Three.js
• Sample floor plans for quick testing
• Georgia Tech themed interface

Requirements

• Python 3.7+
• GCC compiler (for C code)
• Flask and Pillow

Installation

1. Clone the repository:

git clone https://github.com/MengpingZhang/CS6010-Project-Group4.git
cd CS6010-Project-Group4/Code

2. Install Python dependencies:

pip install flask pillow

3. Compile the C program:

Mac/Linux:

gcc -o 3D_Reconstruction 3D_Reconstruction_CSE6010_1.c -lm -fopenmp -O2

Windows:

gcc -o 3D_Reconstruction.exe 3D_Reconstruction_CSE6010_1.c -lm -fopenmp -O2

Usage

1. Start the Flask server:

python app.py

2. Open your browser and go to:

http://127.0.0.1:8000

3. Upload a floor plan image or click a sample image to generate a 3D model

4. Use the toggle buttons to switch between "No Roof" and "With Roof" views

5. Interact with the 3D model:

   • Left Click + Drag: Rotate
   • Right Click + Drag: Pan
   • Scroll: Zoom

Project Structure

Code/
├── 3D_Reconstruction_CSE6010_1.c    # Main C program
├── app.py                            # Flask web server
├── static/
│   ├── app.js                        # Frontend JavaScript
│   ├── style.css                     # Styling
│   └── sample_images/                # Sample floor plans
└── templates/
    └── index.html                    # Main HTML page

Algorithm Overview

1. Image Preprocessing: Otsu thresholding for wall detection
2. Segmentation: BFS flood fill and connected component labeling
3. Door Detection:
   • Red pixel detection for door markers
   • RANSAC-based arc detection for curved doors
4. 3D Extrusion: Wall and floor mesh generation
5. OBJ Export: Standard 3D model format

Technical Details

• Language: C (core algorithm), Python (web server), JavaScript (visualization)
• Parallelization: OpenMP for performance optimization
• Complexity: O(N) where N = number of pixels
• Output Format: Wavefront OBJ files

Team

• Zhang, Mengping
• Ma, Hsu, Chieh
• Gao, Tianxiang
• Nation, Ryan T
• Imran Aziz

Course

CSE 6010 - Computational Problem Solving
Georgia Institute of Technology

License

This project is developed for academic purposes.