High-speed facial recognition system using FaceNet, optimized for real-time performance on low-spec hardware (CPU optimized).
FaceGuard-Pro is an advanced biometric security solution designed for government buildings, airports, and universities. It replaces traditional ID cards with a high-accuracy facial recognition engine.
The standout feature of this project is its efficiency; it delivers lightning-fast recognition even on standard office PCs with weak processors, without requiring a dedicated GPU.
- CPU-Friendly: Highly optimized feature extraction using FaceNet embeddings, making it perfect for low-resource environments.
- Instant Verification: Millisecond-level matching against large databases.
- Universal Camera Integration: Works with any standard USB webcam or IP-based CCTV system.
- Privacy Focused: The system does not store images; it converts faces into encrypted 128-d/512-d numerical embeddings.
- Secure Access Logs: Automatic logging of entry/exit times with identity verification.
| Feature | Interface Preview | Technical Description |
|---|---|---|
| Real-time Recognition |  |
Detecting and identifying faces instantly from a live camera stream. |
| Enrollment System |  |
High-speed facial registration to the encrypted biometric database. |
| Activity Reports |  |
Detailed access logs for security audits and attendance tracking. |
To get the system running, you need to download the pre-trained FaceNet weights:
- Download the Model: Click here to download FaceNet Weights (facenet_keras.h5)
- Directory: Create a folder named
modelsin your project root directory. - Placement: Move the downloaded file into the
models/folder.
Note: This specific model file is optimized for our CPU-friendly inference scripts, ensuring high performance on lower-end hardware.
- Face Localization: MTCNN / Haar Cascades.
- Feature Extraction: FaceNet (Inception ResNet v1).
- Optimization: Image resizing and grayscale pre-processing to maintain high FPS on weak CPUs.
- Database: Secure local storage of identity embeddings.
- Clone the Repo:
git clone [https://github.com/labsisouleimen/FaceGuard-Pro.git](https://github.com/labsisouleimen/FaceGuard-Pro.git)
Identity data is stored as mathematical vectors (embeddings). Even if the database is accessed, the original faces cannot be reconstructed, providing a high layer of security and privacy compliance.
Developed by Souleimen Labsi