e-candeloro/Driver-State-Detection

Detects five distinct driver states (normal, tired, asleep, looking away, distracted) by computing Eye Aspect Ratio, gaze direction, and head pose from 478 Mediapipe face keypoints and iris landmarks. Uses OpenCV's solvePnP for 3D head pose estimation against a canonical face model, then applies rolling PERCLOS (percentage of eyelid closure over time) with decay-based temporal smoothing to reduce false positives. Includes configurable thresholds via command-line arguments for production adaptation across different vehicle setups.

143 stars. No commits in the last 6 months.