A non-intrusive vision system was developed to monitor pilots, including four near infrared (NIR) cameras and four custom-made illuminators. The multi-camera system was installed and tested in a cockpit simulator. State-of-the-art computer vision and machine learning techniques were implemented to extract behavioral features linked to drowsiness in real-time. The key features extracted by the vision system are 3D head pose, blink rate, blink duration, percentage of eye closure (PERCLOS), and mean capped eyelid closure over time (MCEC).
