System Architecture & Anti-Spoofing Pipeline
The Facial Recognition system is engineered to run at high frame rates (25+ FPS) on standard webcams and edge devices such as Raspberry Pi. It incorporates a multi-stage verification pipeline to ensure registered identity authenticity.
Liveness Anti-Spoofing Engine
Protects against photo/video print attacks using eye blink Eye Aspect Ratio (EAR) tracking, frame-to-frame motion variance, and spatial texture artifact analysis before triggering model inference.
MobileNetV2 Classification Engine
Uses lightweight MobileNetV2 transfer learning trained on augmented facial datasets. Includes confidence scoring thresholding (0.75+) to reject unrecognized visitors.
System Execution Pipeline:
[Live Camera Stream (640x480)] ──► [OpenCV Face Detector (Haar Cascade)]
│
▼
[Liveness Verification Pipeline]
• Eye Aspect Ratio (Blink EAR)
• Motion Vector & Texture Check
│
(PASS Liveness)
▼
[MobileNetV2 Facial Feature Extractor]
│
▼
[Softmax Classifier & Confidence Check]
│
(Confidence >= 0.75)
▼
[SQLite Database & Attendance Logger]
• Log timestamped entry to CSV/Excel
• 30-sec duplicate suppression filter
Key Capabilities
- Real-time High FPS Detection: 25-30 FPS live performance with minimal latency (15-25ms detection, 30-50ms recognition).
- Anti-Spoofing Liveness Verification: Prevents photo and smartphone screen replay attacks using eye aspect ratio tracking and texture analysis.
- Automatic Data Augmentation: Automatically enhances user registration images with random rotation (±15°), contrast variation, and brightness jitter.
- Automated Attendance Tracking: Exports timestamped attendance logs to CSV and Excel formats with duplicate filtering (30s window).
- Raspberry Pi Optimization: ARM-optimized TensorFlow builds supporting 640x480 resolution streams at low power draw.
Project Directory Structure
Facial-recognition-system/
├── data/ # Face image datasets & SQLite database
│ └── face_database.db
├── models/ # Saved MobileNetV2 Keras models (.h5)
│ └── face_recognition_model.h5
├── src/ # Python modular backend
│ ├── gui_app.py # Desktop Tkinter GUI application
│ ├── face_detection.py # OpenCV face detection module
│ ├── face_recognition.py # MobileNetV2 transfer learning engine
│ ├── liveness_detection.py # EAR blink tracking anti-spoofing
│ ├── attendance_logger.py # CSV/Excel attendance exporter
│ └── performance_logger.py # Real-time FPS & timing metrics
└── config.json # Model parameters & thresholds