Head Pose Estimation Using Convolutional Neural Network

Abstract

Estimating the head pose is an important capability of a robot when interacting with humans. But there are many difficulties of human head pose estimation, such as extreme pose, lighting, and occlusion, has historically hampered. This paper addresses the problem of head pose estimation with two degrees of freedom (pitch and yaw) using a low-resolution image. We propose a method that uses convolutional neural networks for training and classifying various head poses over a wide range of angles from a single image. Evaluations on public head pose database, Prima database, demonstrate that our method achieves better results than the state-of-the-art.