Projective geometry methodology for head pose estimation using adaboost-LBP

Head pose estimation is a crucial and challenging mission in video-surveillance, human computer interaction (HCI) and visual focus of attention (VFOA). On missing depth information this procedure turns particularly complicated. As core fraction of surveillance, which is fundamental to simplify head pose estimation for high-efficiency production in order to construct a real time system. In this paper, a novel automatic real time head pose estimation system is proposed. It is defined as a hybrid projective geometry method, which presents an approach to reconstruct the depth information in related to the projective geometry information to solve the problem of depth information missing. Camera calibration, pose information and large training sets are not necessary. That is enough with face, eye, and nose information in the projective geometrical model on basis of AdaBoost-LBP. Ultimately, the presented method is experienced on some public datasets and real time webcam video data. It performs competitively with time cost 0.02 to 0.03s and mean error about 1° comparing with state of the art approaches, without additional hardware, supervision and training pose data.