Automatic head pose estimation from a single camera using projective geometry

Abstract

Estimation of human head position and orientation has become an increasingly important issue in human-computer interaction field. Over the last decade, many approaches have been introduced to achieve head pose estimation in both academical and industrial fields, but the low-cost and real-time application still proves to be a difficult task. Motivated by the past researches, we propose an automatic and monocular head pose estimation system. We applied a number of improvements to a direct linear transformation algorithm called Pose from Orthography and Scaling with ITerations (POSIT) and applied it for the head pose estimation. User's virtual head model is also recovered by analyzing laser scan database and corrected by the head pose. The entire process is completely automatic with no need for users to pre-register for identification or initialize their head position. Experiments on a public dataset show realtime performance with lower errors than previous head pose estimation methods.