This project is an academic implementation of a real-time facial recognition system using CNN-based facial embeddings.
It was developed as a final-year engineering project and focuses on edge AI deployment, low-latency processing, and system-level design rather than proposing a new algorithm.
The system was deployed on Raspberry Pi hardware and designed to reflect constraints similar to smart wearable devices such as smart glasses.
- Real-time facial recognition using CNN-based embeddings
- Zero-Lag asynchronous architecture using multithreading
- Edge deployment on Raspberry Pi
- Liveness detection to prevent photo spoofing
- Offline-first cloud synchronization (Google Sheets)
- Physical feedback using a 16×2 LCD display
- Visitor/unknown detection and logging
The system is divided into independent execution paths:
- Video Capture Thread – continuous camera feed
- AI Processing Thread – face recognition and liveness detection
- Network Thread – cloud synchronization
- UI Thread – user interaction and display
This architecture prevents UI freezing on low-power devices.
- Python 3
- OpenCV
- face_recognition (CNN-based embeddings)
- Raspberry Pi
- Tkinter (GUI)
- Google Sheets API
- Multithreading
The system logs attendance events in real time to Google Sheets using OAuth-based authentication. Each successful face recognition triggers an API call that appends a new row containing:
- Student ID
- Student Name
- Date
- Time
This sheet acts as a lightweight cloud backend for attendance monitoring and analysis.
Note:
The displayed sheet represents raw event-level data.
Higher-level attendance summaries (first entry per day, dashboards) can be derived from this data.
This project is an academic prototype developed for learning and demonstration purposes.
The title aligns with existing research literature, but this work does not reproduce or extend any specific published paper.
- Porting the pipeline to wearable smart-glass hardware
- Optimizing inference using TensorFlow Lite
- Adding AR-based output interfaces