Head Pose Estimation from Depth Map

Abstract

In this paper we propose a new method for head pose estimation using the depth sensor Kinect. Our approach infers the head pose based on the symmetry or asymmetry computed on the depth map of the face. This approach does not require the location of the nose or any other feature on the face such has been done in many works, but uses only the depth map of the face. Two features are proposed for characterizing the pan, roll and tilt rotation of the head. The first one, measures the area of nearest region of the face relatively to the face area. The second one, it concerns the symmetry on the depth map of the face. Experiments are conducted on our acquired data. The obtained results are promising and demonstrate the useful of the proposed features.