Head Pose Estimation Using Deep Architectures

Head pose estimation is an important step in many applications related to face analysis, human-computer interaction or activity and behavoiur analysis. We investigate the problem of head pose estimation under a Deep Convolutional Neural Networks (DCNN) framework. Making use of the recent advances in the mentioned field, we test different DCNN architectures and thoroughly evaluate them on UPNA database. We show that adding additional information about location of face fiducial points can increase the accuracy of continuous head pose estimation.