基于三点透视的人脸姿态估计算法  

Face pose estimation based on 3 points perspective

作　　者：丁卓[1] 童卫青[1] 赵骥[1] 

机构地区：[1]华东师范大学计算机科学技术系,上海200241

出　　处：《华东师范大学学报（自然科学版）》2014年第4期43-53,61,共12页Journal of East China Normal University(Natural Science)

摘　　要：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.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.

关 键 词：人脸姿态估计 人脸统计模型 透视变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]