Raspberry Pi Drowsiness Detector Build

I’m putting together a lightweight, always-on driver-drowsiness monitor that runs reliably on a Raspberry Pi Zero 2 W (it should also scale fine to a Pi 4). The system will watch two USB webcams—one on the driver, one on the road—and continuously score fatigue in real time.

Core workflow
• Capture synced frames from the driver-facing and road-facing cameras at a modest, configurable FPS.
• Detect eyes (and ideally mouth & basic head pitch) with whichever method performs fastest on the Zero 2 W—OpenCV Haar, MediaPipe-lite, or a hybrid is fine; I’m not tied to a specific library as long as it stays lean.
• Derive blinks per minute and PERCLOS over a 30–60 s sliding window, then combine them into a 0-100 Risk Score.
• When the score crosses the configurable threshold, pulse a GPIO-driven buzzer (my preferred alert) while enforcing an ~8 s cooldown. A physical mute button or simple keypress should pause alerts for 30 s.
• Overlay a minimalist HUD in one open window showing live PERCLOS, BPM, and score.
• Log every cycle to CSV: timestamp, blink/min, PERCLOS, score, alert flag.
• Pull all tunables—camera indices, FPS, thresholds, GPIO pin numbers, cooldowns—from an external JSON config so I can tweak without editing code.

Please keep the footprint small (Python 3 on Bullseye Lite, headless-friendly), avoid heavyweight GUI frameworks, and note any extra packages I’ll need. I’d like clear installation steps and a short test video or screenshots proving the buzzer fires on simulated drowsiness.

Deliverables
1. Well-commented source code and JSON template
2. Step-by-step setup guide for a fresh Pi Zero 2 W image
3. Sample CSV log and HUD proof-of-life capture
4. Confirmation that the buzzer alert, mute logic, and cooldown work as specced

If this sounds up your alley, let’s talk timing and any clarifications you need.