DriveGuard

Facial-landmark fatigue detection

DriveGuard processes webcam frames to identify sustained eye closure and trigger audible/visual alerts. Python, OpenCV, Dlib, imutils and Pygame connect camera input, facial geometry, temporal state and feedback. The detector distinguishes brief blinks from closure persisting across consecutive frames.

A 68-point landmark predictor localizes the eyes. The work is associated with the IEEE ASIANCON 2025 publication and Dr. Saraswati Patil.

Frame to alert

1. Webcam frame
2. Face & landmarks
3. Eye aspect ratio
4. Frame counter
5. Audio / visual alert

Geometry supplies the frame-level signal; consecutive-frame logic supplies the temporal decision.

Temporal decision logic

• Read and prepare each camera frame, then locate the face and its landmark coordinates.
• Calculate the eye geometry and compare its ratio with the configured threshold.
• Increment the consecutive-frame counter while closure persists; reset it when the open-eye condition returns.
• Trigger the on-screen warning and audio alarm once the sustained condition reaches the required frame count.

Engineering decisions

For six eye landmarks, EAR = (‖p2−p6‖ + ‖p3−p5‖) / (2‖p1−p4‖). Vertical landmark separation is normalized by horizontal eye width. A low EAR increments the closure counter; reopening the eyes resets the sequence. Requiring consecutive low-ratio frames reduces alerts from normal blinks.

The running frame view makes detection state observable and Pygame provides audible feedback once the configured persistence threshold is crossed. Lighting and occlusion affect landmark visibility and therefore the geometric signal. Frame-level state supports inspection of those conditions. Geometry and temporal persistence are distinct stages: a closed-eye frame supplies a measurement, while the counter determines whether closure has continued long enough to alert.

Publication & collaboration

An Intelligent Real-Time Drowsiness Detection System Using Computer Vision And Facial Landmark Analysis For Enhanced Road Safety

IEEE ASIANCON · 2025 · Faculty collaboration with Dr. Saraswati Patil ↗