🎭 Age & Gender Detection

Real-time age and gender detection using OpenCV's Deep Neural Networks (DNN) module and pre-trained Caffe models.

📋 Description

This project detects faces in real-time from webcam feed and estimates:

• Age range (8 categories: 0-2, 4-6, 8-12, 15-20, 25-32, 38-43, 48-53, 60-100)
• Gender (Male/Female)

Built as a personal learning project to explore computer vision and deep learning deployment with OpenCV.

✨ Features

• ✅ Real-time face detection
• ✅ Age estimation (8 age ranges)
• ✅ Gender classification
• ✅ Clean OOP architecture
• ✅ Color-coded bounding boxes (blue for male, pink for female)
• ✅ Confidence scores display

🛠️ Technologies

• Python 3.8+
• OpenCV (cv2) - Computer Vision and DNN module
• NumPy - Array operations
• Pre-trained Caffe models:
  • Face Detection: SSD (Single Shot Detector)
  • Age Estimation: CNN trained on Adience dataset
  • Gender Classification: CNN trained on Adience dataset

📦 Installation

1. Clone the repository

git clone https://github.com/Fanfulla/cv_webcam.git
cd cv_webcam

2. Create virtual environment

python3 -m venv venv
source venv/bin/activate  # On macOS/Linux

3. Install dependencies

pip install opencv-contrib-python numpy

4. Download pre-trained models

The models are already included in the models/ directory:

• Face Detection: deploy.prototxt, res10_300x300_ssd_iter_140000.caffemodel
• Age Estimation: age_deploy.prototxt, age_net.caffemodel
• Gender Classification: gender_deploy.prototxt, gender_net.caffemodel

🚀 Usage

python3 main.py

• The webcam window will open
• Face detection and age/gender estimation run in real-time
• Press 'q' or ESC to exit

📁 Project Structure

cv_webcam/
├── main.py                 # Entry point
├── face_detector.py        # Face detection class
├── age_estimator.py        # Age estimation class
├── gender_estimator.py     # Gender classification class
├── video_processor.py      # Webcam and video processing
├── models/                 # Pre-trained models
│   ├── deploy.prototxt
│   ├── res10_300x300_ssd_iter_140000.caffemodel
│   ├── age_deploy.prototxt
│   ├── age_net.caffemodel
│   ├── gender_deploy.prototxt
│   └── gender_net.caffemodel
├── requirements.txt        # Python dependencies
└── README.md

🧠 How It Works

1. Face Detection: Uses OpenCV's DNN module with a pre-trained SSD model to detect faces
2. Face Extraction: Crops detected faces from the frame
3. Preprocessing: Resizes faces to 227x227 and applies mean subtraction
4. Age/Gender Prediction: Passes preprocessed faces through CNNs
5. Visualization: Draws bounding boxes and labels on the original frame

🎓 Learning Outcomes

This project helped me learn:

• Object-oriented programming in Python
• OpenCV DNN module and pre-trained model deployment
• Real-time video processing
• Image preprocessing (blob creation, mean subtraction)
• NumPy array operations and image slicing
• Git version control workflow

📊 Model Information

Age Ranges

The model classifies faces into 8 age ranges:

• (0-2), (4-6), (8-12), (15-20), (25-32), (38-43), (48-53), (60-100)

Accuracy

These pre-trained models provide reasonable accuracy for general use but may vary based on:

• Lighting conditions
• Face angle
• Image quality
• Ethnicity representation in training data

⚠️ Limitations

• Age estimation returns ranges, not exact ages
• Performance depends on lighting and camera quality
• Models may have bias based on training data
• Not suitable for production/critical applications

🔮 Future Improvements

• Add emotion detection
• Save screenshots on keypress
• Support for image/video file input
• Performance metrics and FPS display
• Model fine-tuning on custom datasets
• Docker deployment

📚 Resources & Credits

• OpenCV DNN Module: OpenCV Documentation
• Age/Gender Models: Based on research by Gil Levi and Tal Hassner
• Face Detection Model: OpenCV's pre-trained SSD model
• Dataset: Adience Benchmark

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

👤 Author

Salvatore (Fanfulla)

• GitHub: @Fanfulla

🙏 Acknowledgments

• Thanks to the OpenCV community for excellent documentation
• Gil Levi and Tal Hassner for the age/gender models
• Profession.AI for the learning opportunity

---

Note: This is a personal learning project. Models and predictions should not be used for critical decision-making.