基于仿射变换和线性回归的3D人脸姿态估计方法

3D face pose estimation based on affine transformation and linear regression

提出了一种由仿射变换关系到线性回归的3D人脸空间姿态估计方法。即跟踪到人脸特征点后,根据仿射变换关系得到人脸姿态的粗估计值,以这个粗估计值作为人脸姿态的初始值,再通过线性回归迭代求得人脸姿态的精确值。实验结果表明,该方法在较大的姿态变化范围内,具有良好的估计精确度和鲁棒性。

A method for estimating 3D space face pose was proposed based on affine transformation and linear regression. Namely, after tracking face feature points, a roughly estimated value was obtained based on affine transformation relationship, and then this rough value was taken as the starting value, precise value of face pose was obtained based on linear regression iteration. Finally, the experimental results show that the method achieves better estimation accuracy and robustness in a wide range of face poses.