Recently, deep learning approaches have achieved promising results in variousfields of computer vision. In this paper, we tackle the problem of head poseestimation through a Convolutional Neural Network (CNN). Differently from otherproposals in the literature, the described system is able to work directly andbased only on raw depth data. Moreover, the head pose estimation is solved as aregression problem and does not rely on visual facial features like faciallandmarks. We tested our system on a well known public dataset, Biwi KinectHead Pose, showing that our approach achieves state-of-art results and is ableto meet real time performance requirements.
展开▼