Raspberry Pi Driver Monitoring System

This project provides a Raspberry Pi oriented driver monitoring server built with Flask, OpenCV, MediaPipe model assets, and GPIO-based alerts.

It accepts uploaded JPEG frames, detects face and attention state, and exposes both a live MJPEG stream and JSON status output. The current runtime classifies:

• NO_FACE
• NORMAL
• DISTRACTED
• DROWSY

It also supports:

• LED and buzzer alerts through Raspberry Pi GPIO
• looped melodies for sustained DISTRACTED and DROWSY states
• auto-download of required OpenCV cascade XML files if missing
• systemd deployment for headless startup on boot

Repository Structure

• pi_server.py: main Flask server and processing loop
• models/face_landmarker.task: local face landmark model asset
• cascades/: OpenCV cascade storage/download directory
• driver_monitor.service: example systemd unit

Requirements

• Python 3.10+
• Raspberry Pi OS or another Linux environment
• Camera client that posts JPEG frames to the server
• Optional Raspberry Pi GPIO hardware for LED/buzzer output

Install Python dependencies:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Running Locally

Start the server:

python3 pi_server.py

Default server address:

• http://0.0.0.0:5000/

Useful endpoints:

• GET / : simple monitoring page
• GET /status : JSON status and last frame age
• GET /stream : MJPEG video stream
• POST /frame : upload a frame under multipart field frame

Example frame upload:

curl -X POST http://127.0.0.1:5000/frame \
  -F "frame=@sample.jpg"

Example status check:

curl http://127.0.0.1:5000/status

GPIO Behavior

Current output logic:

• NO_FACE: LED on, buzzer off
• NORMAL: LED off, buzzer off
• DISTRACTED: after 3 seconds, starts looping Fur Elise
• DROWSY: after 3 seconds, starts looping Super Mario; after 6 seconds total, LED also turns on

If GPIO initialization fails, the app falls back to a mock backend and continues running without hardware control.

Deployment With systemd

An example service file is included as driver_monitor.service.

Before enabling it, update paths if your installation directory differs from /home/pi/driver-monitor.

Typical setup:

sudo cp driver_monitor.service /etc/systemd/system/driver_monitor.service
sudo systemctl daemon-reload
sudo systemctl enable driver_monitor.service
sudo systemctl start driver_monitor.service
sudo systemctl status driver_monitor.service

Notes

• pi_server copy.py is a local backup copy and is intentionally excluded from version control.
• The project expects models/face_landmarker.task to be present locally.
• Haar cascade files are downloaded automatically if OpenCV cannot find them on the system.