基于深度值和M-估计的人脸姿态估计  

Face Pose Estimation Based on Depth and M-estimation

作　　者：邱丽梅[1,2] 邱思杰 QIU Li-mei;QIU Si-jie(School of Mechanical&Electronic Engineering,Sanming University,Sanming 365004,China;Fujian University Engineering Research Center of Modern Mechanical Design and Manufacturing Technology,Sanming 365004,China)

机构地区：[1]三明学院机电工程学院,福建三明365004 [2]机械现代设计制造技术福建省高校工程研究中心,福建三明365004

出　　处：《三明学院学报》2018年第2期49-55,共7页Journal of Sanming University

基　　金：福建省自然科学基金计划项目(2017J01778);福建省科技项目(JK2015045);三明市科技项目(2016-G-8);福建省科技厅科技项目(2014H0004)

摘　　要：对于单幅人脸图像姿态估计,针对现有的估计方法大多存在缺乏人脸特征深度信息而造成姿态估计过程病态化的缺陷,及最小二乘法不能很好地区分内外特征点,从而影响估计精度和鲁棒性的问题。提出利用透视原理和迭代算法计算人脸特征点的深度值,在最小二乘法的基础上,加入M-估计优化算法,精确估计出人脸三维空间姿态。实验结果表明,该方法不仅克服了姿态估计的病态化问题,而且具有良好的姿态估计精度和鲁棒性。Most of the existing pose estimation methods have the defect that the process of pose estimation is ill from lack of the depth information,and cannot distinguish the internal and external points of the characteristics because of the least square method,affecting the accuracy and robustness of estimation.Therefore,a new approach of using the principle of perspective and iterative algorithm to calculate the depth of facial feature points is proposed in this paper for estimating 3D face pose from a single image.Then,the M-estimation algorithm is added on the basis of the least square method and the 3D face pose is estimated accurately.Experimental results show that this method not only overcomes the ill posed problem of pose estimation,but also has good accuracy and robustness of pose estimation.

关 键 词：人脸姿态估计 深度值 特征点 最小二乘法 M-估计算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]