Detection of Face Orientation by 3D Measurement of Facial Features

抄録

An approach to detect human facial orientation by reconstructing the three-dimensional position of facial points from stereo images is proposed. The purpose is to develop a flexible and accurate facial orientation detection method for human interaction with computers. Four facial points (pupils and mouth corners) are extracted using a global-to-local method and their three-dimensional coordinates are reconstructed from stereo images. Facial orientation is obtained by calculating the normal direction of the plane formed by three facial feature points. The advantages of the proposed method are first, that it is flexible and easy to adapt to different persons and circumstances; and second, that it eliminates the need for prior knowledge of image vs. model correspondence or the physical parameters of facial features. Analytical and experimental results show that the proposed method performs well when the head is not far away from the frontal view. These characteristics are especially useful for constructing a human-computer interaction system.