Smart Attendance System – Raspberry Pi 4

Offline, portable face-recognition-based attendance system.

Students press a button, blink, and get marked present.
No internet connection required.

---

What Makes This Different

Most attendance systems require:

• Active student participation (QR scanning, manual marking)
• Internet connectivity
• Expensive hardware
• IT support for deployment

This system:

• Passively verifies presence through face recognition + liveness detection
• Works completely offline (local database, no cloud)
• Costs less (Raspberry Pi 4 + webcam + components)
• Student-operated (press button → blink → done)
• Teacher controls remotely via web dashboard

---

How It Works

Student Flow (5–7 seconds)

1. Press button on device
2. Look at camera
3. Blink when prompted (liveness check)
4. LCD shows: "John Doe - Marked Present!"
5. Pass device to next student

Teacher Flow

• Start session from web dashboard (laptop/phone)
• Students mark themselves using the device
• View live attendance count
• End session and download CSV report

---

Core Architecture

Camera → Face Detection → Embedding Extraction → Cosine Matching
                              ↓
                        Blink Liveness Check
                              ↓
                       Attendance Engine
                              ↓
                         PostgreSQL
                              ↓
                        FastAPI Backend

---

Technology Stack

Core Recognition:

• InsightFace (buffalo_sc model) – CPU-optimized face recognition
• ONNX Runtime (CPU execution)
• dlib – Blink detection via Eye Aspect Ratio (EAR)
• OpenCV – Frame processing

Backend:

• FastAPI – REST API + teacher dashboard
• PostgreSQL – Student data, attendance records, sessions
• SQLAlchemy ORM

Hardware (Raspberry Pi):

• GPIO – Button input, LED status indicators
• I2C LCD (16×2) – Student feedback messages
• USB Webcam (720p minimum)

Development:

• Python 3.10
• Tested on Windows (development), deployed on Raspberry Pi OS

---

Target Performance

• 5-7 seconds per student
• 80 students processed in under 15 minutes
• Fully offline operation
• Designed for Raspberry Pi 4 (CPU-only execution)

---

In Progress

• Anti-spoofing enhancement (currently: blink + face verification)
• Email/SMS notifications
• Hardware integration (GPIO, LCD, LEDs)

Anti-Spoofing Strategy

Current Implementation:

1. Blink detection – Defeats printed photos
2. Face verification during blink – Prevents swap attacks (showing photo → person blinks)

Known Limitations:

• Pre-recorded videos with blinks may pass (low probability in practice)
• Live video calls on screens are not fully prevented

Exploring:

• Challenge-response system (random actions: smile, turn head)
• Deep learning models (deferred due to Raspberry Pi constraints)

---

Contributing

Currently under active development. Suggestions and feedback are welcome, especially on:

• Anti-spoofing techniques suitable for Raspberry Pi
• UI/UX improvements for teacher dashboard