Loading face detection model...

Ready Alert

Real-time driver drowsiness detection — no ML model, just math

Live Feed

Loading...
EAR: —

Eye Aspect Ratio Over Time

How It Works

1
Calibrate — 3 seconds captures your personal alert EAR baseline while you look normally at the camera.
2
Detect — MediaPipe extracts 468 facial landmarks per frame. Your EAR is compared to your personal threshold across a 30-frame rolling window.
3
Alert — If ≥48% of frames in the window fall below your threshold, a drowsy alert fires with an audio beep.

Parameters (Grid-Searched)

0.79
EAR threshold
(ratio of personal baseline)
48%
Alert percentage
(of 4-second window)
25
Subjects in parameter sweep
0
ML models at inference time