GHS Hazard Label Detection Application

A real-time application for detecting and tracking Globally Harmonized System (GHS) hazard labels using computer vision.

Features

• Real-time Detection: Identify GHS hazard labels in camera feeds with YOLO-based detection
• Object Tracking: Track detected hazard labels across video frames
• Multi-Platform: Works on Windows, macOS, and Linux
• Camera Support: Compatible with webcams, IP cameras, and video files
• Export Capabilities: Save screenshots and detection results to CSV

Installation

1. Clone the repository:

git clone https://github.com/yourusername/hazard-label-dataset.git
cd hazard-label-dataset

2. Create a virtual environment (optional but recommended):

python -m venv venv
source venv/bin/activate  # On Windows: .\venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

Running the Application

The application requires a YOLO model for detection. You can use the provided script to download a sample model:

python download_sample_model.py

Start the Application

python src/app.py

Using the Detection Interface

1. Camera Controls:

   • Select a camera from the dropdown menu
   • Choose resolution and FPS settings
   • Click "Start Camera" to begin the video feed

2. Model Selection:

   • Select a model from the dropdown or use "Load Model From File..."
   • The application will look for models in the models/ directory

3. Detection Settings:

   • Adjust confidence threshold to control detection sensitivity
   • Modify IoU threshold for overlapping detection handling
   • Toggle tracking on/off as needed

4. Results and Export:

   • View detected hazard labels in the results table
   • Take screenshots or export detection data using the File menu

Supported GHS Hazard Label Classes

The application can detect the following GHS hazard pictograms (dependent on your trained model):

• Explosive (GHS01)
• Flammable (GHS02)
• Oxidizing (GHS03)
• Compressed Gas (GHS04)
• Corrosive (GHS05)
• Toxic (GHS06)
• Harmful/Irritant (GHS07)
• Health Hazard (GHS08)
• Environmental Hazard (GHS09)

Troubleshooting

Camera Issues

• If your camera doesn't appear in the list, click the refresh button
• Try different backend options in the dropdown
• For IP cameras, select "IP/URL Camera" and enter the RTSP/HTTP URL

Detection Problems

• Ensure proper lighting for better label recognition
• Adjust the confidence threshold slider if detection is too sensitive/not sensitive enough
• Try different model files if certain hazard types aren't being detected

Development

This project uses:

• PyQt6 for the user interface
• Ultralytics YOLOv8 for object detection
• Supervision for tracking and visualization

Project Structure

hazard-label-dataset/
├── models/               # YOLO model files (.pt)
├── output/               # Screenshots and exported data
├── src/
│   ├── app/
│   │   ├── controllers/  # Application logic
│   │   ├── models/       # Data handling
│   │   ├── utils/        # Helper utilities
│   │   └── views/        # UI components
│   └── app.py            # Entry point
├── requirements.txt      # Dependencies
└── download_sample_model.py  # Utility script

License

MIT License