Head pose detection with one camera based on pupil and nostril detection technique

Abstract

A remote video-based head pose detection system is useful to observe the behavior of the driver, e.g., prevention oft the driver from failing to keep the eyes on the road. In addition, if the precision of the system is high, it may be used as a pointing device. We have already proposed a horizontal and vertical head pose estimation method using the 3-D positions of the two pupils and two nostrils, which are obtained from the stereo cameras. In the method, the horizontal and vertical angles of the average vector of the normal vector of the plane including the two pupils and one nostril and that of the plane including the two pupils and another nostril were determined as the head angle. The system showed the measurable ranges of more than plusmn 40 deg in horizontal and 30 deg in vertical. In this paper, we propose the head pose detection method with one camera in order to increase the horizontal range. The 3D positions of the pupils and nostrils are determined by giving the distances among the feature points as restraint conditions. The experimental results showed the measurable ranges of approx. plusmn 45 deg in horizontal and 40 deg in vertical. In addition, the one-camera method indicated high resolution compared to our previous stereo method. Finally, the proposed one-camera method was introduced into each of the two stereo-calibrated cameras, placed with 30 deg separated. The two-camera system showed a wide horizontal range of plusmn 60 deg.