Automated Construction Progress Monitoring

This project allows for automated monitoring of construction sites using drone photogrammetry and 3D reconstruction. This system integrates advanced computer vision techniques to provide real-time insights into civil engineering projects.

Features

• 3D Site Model: View high-fidelity current and previous 3D scans of the construction site.
• Progress Monitoring: Track overall completion against the planned schedule with precise metrics.
• Heatmaps: Visualize construction activity zones to identify bottlenecks.
• Automated Alerts: Receive real-time notifications for potential schedule delays or safety issues.
• Crack Detection: Automated analysis of surface defects using deep learning.

Execution Flow

The system processes drone imagery through a multi-stage pipeline:

graph TD
    A[Input Drone Imagery] --> B["Structure from Motion (SfM)"]
    B --> C[Sparse Point Cloud Generation]
    C --> D["Multi-View Stereo (AA-RMVSNet)"]
    D --> E[Depth Map Estimation]
    E --> F[Point Cloud Fusion]
    F --> G[3D Site Model]
    A --> H[Crack Detection Model]
    H --> I[Defect Analysis Report]
    G --> J[Progress Visualization]
    I --> J

Loading

Project Structure

The project is organized into frontend visualization and backend processing components:

Project_Devs/
├── backend/                  # 3D reconstruction and analysis algorithms (Python)
│   ├── datasets/             # Data loading and preprocessing modules
│   ├── models/               # PyTorch network architectures (AA-RMVSNet)
│   ├── sfm/                  # Structure from Motion integration utilities
│   ├── eval.py               # Main script for depth estimation and evaluation
│   ├── fusion.py             # Script for fusing depth maps into point clouds
│   └── utils.py              # General utility functions for I/O and processing
├── models/                   # Pretrained machine learning models
│   ├── aa_rmvsnet.pth        # Weights for the 3D reconstruction network
│   └── crack_detection_model.pth # Weights for the crack detection network
├── src/                      # Frontend user interface application (React)
│   ├── components/           # Reusable UI components (Dashboard, ModelViewer, etc.)
│   ├── pages/                # Main application routes and page layouts
|   └── lib/                  # Utility libraries and constants
├── public/                   # Static assets and entry HTML
└── README.md                 # Project documentation and usage guide

Getting Started

Frontend

1. Install dependencies:

   npm install

2. Run the development server:

   npm run dev

Backend Processing

To use the 3D reconstruction and analysis features, a Python environment (Python 3.6+ recommended) is required.

1. Install Dependencies: It is recommended to use a virtual environment or Conda.

   # Using pip
   pip install -r backend/requirements.txt

   Dependencies include: torch, torchvision, opencv-python, plyfile, tensorboardx, numpy.

2. 3D Reconstruction: Navigate to the backend directory and run the evaluation script:

   cd backend
   python eval.py --testpath <input_data_path> --outdir <output_path> --loadckpt ../models/aa_rmvsnet.pth

   Note: The AA-RMVSNet model is optimized for CUDA-capable GPUs.

3. Crack Detection: The crack detection workflow utilizes the crack_detection_model.pth located in the models/ directory. Ensure input images are high-resolution for optimal defect recognition.

Technologies

• Frontend: React, Vite, Tailwind CSS, Three.js (via react-three-fiber).
• Backend: Python, PyTorch (AA-RMVSNet).

License

[License Information]