DriveGuard Driver Monitoring System

Driver distraction and drowsiness are safety-critical. A camera-based monitor can surface early cues—eye closure, gaze away from the road, head pose changes, and yawning—before they compound into risk.

• Real-time facial landmark analysis
• Multi-signal drowsiness / distraction heuristics
• Temporal filtering to reduce false alarms
• Audible warnings when thresholds are crossed

Webcam frames → MediaPipe Face Mesh / related landmarks → feature extractors (EAR, gaze proxy, head pose, mouth aspect) → temporal warning state machine → audio alert.

Python application combining OpenCV capture with MediaPipe inference. Each cue contributes to a warning score; sustained cues trigger alerts rather than single-frame spikes.

Exercised with staged behaviors (eyes closed, looking aside, yawning) to tune thresholds and temporal windows.

Produced a working real-time monitor with eye closure, gaze, head-position, and yawning cues plus audio alerts.

• Lighting sensitivity
• False positives from brief natural blinks
• Camera placement variability

Temporal logic is essential—raw landmark features are noisy without hysteresis and sustained-evidence rules.

• Better calibration UX
• On-device optimization
• Expanded dataset for threshold tuning