A head posture estimation method based on 3-D images measurement for intuitive human-system interaction

Abstract

Real-time detection of the person's head posture is very important in field such as safe driving of automobile, man-machine interface of game and robot, and so on. Some techniques about the posture estimation of driver's head are existed, like detecting the sight line from the position of the pupil and iris of the eyes, using the positional relationship between the eyes and nose, and so on. However, there is a problem that the result of image measuring is influenced by ambient light and face makeup of the driver in these techniques. And then, it cannot estimate a wide range of face direction. In order to solve this problem, we propose a technique of estimate the head posture of the driver by three-dimensional (3-D) image analysis which is based on the infrared pattern projection technique. The technique we proposed includes two portions: one is 3-D image measurement of driver's head and the other is the head posture estimation by 3-D image analysis. As the 3-D image measurement method, in order to obtain a reliable 3-D image, we use slit pattern projection technique; in order to keep the 3-D measurement accurate of the portion near the nose, and the 3-D measurement speed, we only use three stripes to project on the face; in order to reduce the influence of external light on the people's body, we use infrared pattern projection. To estimate the head posture quickly, we face contour modeled as linear approximation, and then use the relationship between the face degree and the slope of the line to judge the face direction.