Blink Counter – Raspberry Pi Zero 2W

Real-time blink detection system using a Raspberry Pi Zero 2W, Camera v3 (IMX708), TFLite, and MQTT.

Features

• Real-time eye blink detection using a TFLite model with Haar Cascade face/eye detection
• MQTT publishing of blink counts at configurable intervals with auto-reconnect
• Web-based configuration portal (Flask) for Wi-Fi, MQTT, camera, and model settings
• Automatic AP hotspot mode when Wi-Fi is unavailable (accessible at 192.168.4.1:5000)
• Dual-stream Picamera2 pipeline: low-res YUV for fast inference, optional high-res for preview
• Optimized for Pi Zero 2W: minimal CPU/RAM usage in headless mode

Quick Start

1. Copy this folder to your Raspberry Pi Zero 2W
2. Run the setup script:

   cd /path/to/blink-counter
   sudo bash setup.sh

3. Reboot:

   sudo reboot

4. The app starts automatically. Configure via the web portal at http://<pi-ip>:5000

Project Structure

File	Description
TFlite_tester_Copy.py	Main application script
eye_state_model2.tflite	TFLite eye-state classification model
rpi_config.json	Runtime configuration (Wi-Fi, MQTT, camera, thresholds)
setup.sh	One-command installer for all dependencies
blink_app	Bash launcher script (handles venv + root)
check_flask_ap.sh	Diagnostic script for Flask/AP networking
autostart/	Desktop autostart entries

Configuration

Access the web portal at http://<pi-ip>:5000 to configure:

• Wi-Fi: SSID and password
• MQTT: Server, port, credentials, publish interval
• Camera: Resolution, FPS, preview toggle
• Model: Blink thresholds, detection stride, TFLite threads

Dependencies

All installed automatically by setup.sh:

• python3-picamera2, libcamera-apps
• python3-opencv
• python3-flask
• paho-mqtt
• tflite-runtime (from piwheels)

Hardware

• Raspberry Pi Zero 2W
• Raspberry Pi Camera Module v3 (IMX708)

License

MIT