基于三点透视的人脸姿态估计算法

提出了基于特征点透视变换和人脸统计模型的人脸姿态估计算法.首先建立以左右眼睛点和鼻尖为几何特征点的人脸统计模型,然后推导出这三个特征点在图像平面坐标系和三维世界坐标系间的对应关系,进而得到从图像平面上的三个特征点估算人脸在三维空间姿态的理论方程,最后运用迭代方法给出了一个完整的快速人脸姿态估计算法.实验结果表明,本算法精度高,速度快,具有较好的鲁棒性.

Face pose estimation based on 3 points perspective

This paper presents an approach of face pose estimation
which is based on the feature point perspective transformation and
the statistical face model. First, we build the statistic face model
using three geometric feature points (2 eye points and one nose
tip). Second, we deduce the corresponding relationship of the three
points between the image coordination system and the world
coordination system, and obtain the equations for estimating the
face pose from the three points in the face image. Finally, using
iterative method, we give a complete fast algorithm of face pose
estimation. The experiment result shows that the proposed approach
is accurate, fast and robust.