A Real-time System for Head Tracking and Pose Estimation for Driver’s Alertness

Abstract

Driver's visual attention provides important clues about his/ her activities and awareness. To monitor driver's awareness, this paper proposes a real-time
person-independent head tracking and pose estimation system using a monochromatic camera. The tracking and head-pose estimation tasks are formulated as regression problems. Three regression methods are proposed: (i) individual mapping on images for head tracking, (ii) direct mapping to subspace for head tracking, which predicts a subspace from one sample, and (iii) semantic piecewise regression for head-pose estimation. The approaches are evaluated on standard databases, and on several videos collected in vehicle environments.