基于三维模型和仿射对应原理的人脸姿态估计方法

Face Pose Estimation Based on 3D Models and Affine Correspondences

该文提出了一种基于人脸三维模型和仿射对应原理从单目视频图像序列中估计人脸空间姿态的方法.其主要思想是利用人脸的三维模型生成特征点正面平行投影,并估算输入帧和该正面平行投影之间的仿射变换参数,然后根据圆椭圆之间的仿射对应关系得到描述人脸空间姿态的6个参数(3个旋转分量,3个平移分量)的粗略估计值,最后通过基于ICP(IterativeClosestPoints反复最近点)算法的优化迭代过程得到精确值.对石膏像和真实人脸进行的实验结果表明该算法能在较大的姿态变化范围内实现精确的人脸姿态估计.

The paper proposes a model-based method for determining human head poses from a monocular 2D image. The main idea is to use a 3D head model to generate a virtual fronto-parallel view for estimating the rigid motion of the head. The affine correspondence between some features of the virtual view and the input image is firstly established. Then, the rotation and translation parameters are roughly estimated by a circle-ellipse correspondence technique based on the affine parameters. Finally, an Iterative Closest Points (ICP) algorithm is utilized further to refine the results from the above step. Experimental results show that the method can accurately recover head poses in a wide range of head motion.