Drivers’ state estimation by means of robust 3D head pose estimation for enhancing perceptual scene awareness

Reliable and accurate car driver head pose estimation is an important component for the next generation of advanced driver assistance systems in automated vehicles. Such a system should not be invasive nor rely on prior training and calibration, thus being independent of the driver's identity. We report on a highly accurate system we have adapted from graphics for automotive applications and demonstrate its performance on new challenging real-world data. Our system automatically fits a statistical 3D face model to depth measurements of a driver's face, acquired with a low-end sensor. We can demonstrate improvements over state-of-the-art camera-based 2D face tracking approaches. Our system delivers a full 6-DOF pose with very little degradation from strong illumination changes or out-of-plane rotations of more than 50 degree. We discuss potential applications of this approach for enhancing driver’s perceptual awareness under high work load and distraction for different degrees of automated driving based on our new object detectability concept