Abstract: For three-dimensional portrait pose estimation, combined with standard face, a human face statistical knowledge and projective geometry a method is proposed. Firstly, repeatedly select corners of mouth, eyes and points under the nose to get the characteristic triangular about face and build a three-dimensional model; secondly statistics face three-dimensional model parameters; Rotation matrix is then calculated based on the perspective projection model; Finally, calculate the three-dimensional gesture based on least squares method. The method of this article can be implemented without the auxiliary equipment, distinct feature points are easy to be located, with some accuracy.

Key words: pose estimation, photographic superimposition, three-dimensional model, least squares method

摘要： 针对二维人像的三维姿态估计，结合标准脸型、人脸的统计知识和射影几何提出一种方法：多次选取标准脸型眼角点、嘴角点、鼻下点等特征点得到特征三角形并确定人脸三维模型；其次计算人脸三维模型参数；基于透视投影模型计算旋转矩阵；基于最小二乘法计算三维姿态。提取的特征点易于标定，无需任何辅助设备，具有一定的准确性。

关键词: 姿态估计, 颅像重合, 三维模型, 最小二乘法