Real-Time Head Pose Estimation using Reconstructed 3D Face Data from Stereo Image Pair

In this paper, we describe a real-time 3D human head pose estimation algorithm by means of eye tracking and on-line reconstructed 3D face data from a stereo image pair. A frame by frame based face coordinate system is formed using eye locations and a face plane which doesn´t rely on distinctive facial features. The rest of the algorithm follows the conventional approach of pose estimation between the face coordinate system and a reference coordinate system without any initialization and model fitting. By not relying on arbitrary data on the face such as nose ridge line or a mouth corner points, we minimize the impact of errors which may come from their 3D estimates and facial expression changes. Experimental results demonstrate the accuracy and robustness of the algorithm.