A high-performance real-time face recognition application built with MediaPipe, DeepFace, and OpenCV. Optimized for speed with threading to maintain smooth video streaming while performing face detection and recognition in the background.

📹 Demo Video

Watch the full demo: YouTube - Face Recognition Demo

🚀 Features

Real-Time Face Detection: Runs at 60 FPS using MediaPipe's BlazeFace model
Asynchronous Recognition: Non-blocking face identification using threading
Fast Face Matching: ArcFace model for accurate face comparison
Smooth UI: Instant visual feedback with bounding boxes and FPS counter
Unknown Face Detection: Displays "Unknown" for unrecognized faces

🛠️ Technology Stack

Core Libraries

Technology	Purpose
MediaPipe	High-speed face detection using BlazeFace model
DeepFace	Face recognition and matching with ArcFace
OpenCV (cv2)	Video capture and real-time frame processing
TensorFlow	Deep learning framework for neural networks
Keras	High-level neural networks API
Threading	Asynchronous face recognition to prevent video lag

Additional Dependencies

Package	Version	Purpose
NumPy	≥1.14.0	Numerical computing
Pandas	≥0.23.4	Data manipulation
Pillow	≥5.2.0	Image processing
MTCNN	≥0.1.0	Face detection alternative
RetinaFace	≥0.0.14	Enhanced face detection
Flask	≥1.1.2	Web API framework
Flask-CORS	≥4.0.1	Cross-origin resource sharing
Gunicorn	≥20.1.0	WSGI HTTP Server
Python-dotenv	≥1.0.1	Environment variable management

📋 Requirements

Python 3.11 (tested on Python 3.11)
Webcam/Camera for real-time capture
Trained Face Database: Store face images in known_faces/ directory

⚙️ Installation

1. Clone the Repository

git clone https://github.com/ssbg04/facerecog.git
cd facerecog

2. Create Virtual Environment

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

3. Install Dependencies

pip install -r requirements.txt

4. Download MediaPipe Model

The BlazeFace model file (blaze_face_short_range.tflite) needs to be in the project root directory. Download from:

MediaPipe Model Zoo

🏃 Usage

Basic Usage

python app.py

The application will:

Open your webcam
Display a window titled "Smooth Face Recognition"
Detect faces in real-time
Identify known faces from the database
Show FPS counter in the top-left corner

Press Q to quit

Prepare Your Face Database

Create a known_faces/ directory and organize it as follows:

known_faces/
├── person1.jpg
├── person2.jpg
├── john_doe.jpg
└── ...

The system extracts names from filenames (before the file extension).

🎨 How It Works

Architecture

Input Frame
    ↓
[Detection] ← MediaPipe (60 FPS) - Detects face locations
    ↓
[Threading] ← Async Recognition - Doesn't block video
    ↓
[Recognition] ← DeepFace/ArcFace - Identifies face
    ↓
[UI Render] ← OpenCV - Draws bounding box + name
    ↓
Display Output (Smooth Video Stream)

Key Optimizations

High-Speed Detection: MediaPipe's lightweight BlazeFace runs at 60 FPS
Async Processing: Face recognition runs on a separate thread
Early Stopping: Only recognizes when a new face is detected
Efficient Cropping: Uses enforce_detection=False for faster matching

📊 Performance

Detection FPS: ~60 FPS
Recognition Time: ~100-500ms (depends on database size)
Smooth Video: Yes - threading prevents frame drops

🔧 Configuration

Edit the following variables in app.py:

DB_PATH = "known_faces"              # Path to face database
MODEL_PATH = 'blaze_face_short_range.tflite'  # MediaPipe model

🐛 Troubleshooting

Issue	Solution
Model file not found	Download `blaze_face_short_range.tflite` to project root
Webcam not working	Check camera permissions or try `cv2.VideoCapture(1)`
Slow performance	Reduce image resolution or database size
"Unknown" for known faces	Verify image quality and lighting in database

📝 License

No specific license specified. Please check the repository for details.

👤 Author

ssbg04 - GitHub Profile

🤝 Contributing

Feel free to fork, modify, and submit pull requests!

📚 References

Last Updated: June 2026 | Status: Active Development