基于人脸特征点和线性回归的3D人脸姿态估计方法

3D Face Pose Estimation Based on Face Feature Points and Linear Regression

改进传统的活动形状模型法,准确地提取人脸特征点后,利用人脸特征点初步估计人脸姿态,以初步估计值为初始值,通过线性回归迭代算法,精确估计3D人脸空间姿态。实验结果表明,本文提出的新方法不仅可以获得稳定和唯一的3D人脸空间姿态,而且与同类方法比较具有较好的姿态估计精确度。

In this paper, the improved method of active shape model is used to extract face feature points accurately, then a coarse estimation value is obtained by using face feature points, and takes this coarse value as the starting value, precise value of 3D face pose is obtained based on linear regression iteration. The experiment results show that the new method not only can achieve steady and only 3D face pose, but also has better estimation accuracy compared with the same kind of method.