基于改进SURF算法的人脸点云配准  

Face point cloud registration based on improved SURF algorithm

作　　者：郭昱 佘二永[3] 王清华[1] 李振华[1] GUO Yu;She Eryong;WANG Qinghua;LI Zhenhua(School of Science, Nanjing University of Science and Technology, Nanjing 210094, China;Beijing Zen-AI Technology Ltd, Haidian District, BeiJing 100089, China;China National Defense Science and Technology Information Center, BeiJing 100089, China)

机构地区：[1]南京理工大学理学院,南京210094 [2]北京仁光科技有限公司,北京100089 [3]中国国防科技信息中心,北京100089

出　　处：《光学技术》2018年第3期333-338,共6页Optical Technique

摘　　要：精准的三维人脸重建是三维人脸识别、三维人脸表情仿真等技术实现的重要前提。基于以往图像特征点和点云数据的三维配准算法研究,提出了一种计算量小、实时性较高的人脸配准算法。提取人脸图像特征点,计算64维的SURF描述符;利用RANSAC算法剔除不稳定匹配点;利用奇异值分解SVD求解粗配准变换矩阵;利用改进的最近点迭代算法求解最终变换矩阵。实验结果显示配准误差只有8.71895×10^-5m^2,总耗时为6.61s,相比较SIFT算法和手动寻找匹配点,速度快、精度高。The accurate 3 D face reconstruction is an important prerequisite for 3 D face recognition and 3 D facial expression simulation. A face registration algorithm with small computation and high real-time is proposed based on previous research of image feature point and three-dimensional registration algorithm for point cloud data. The feature points of face images are extracted and 64 dimensional SURF descriptors are computed. Using the RANSAC algorithm,the unstable matching points are eliminated.The 2 D image feature matching points are introduced into the 3 D point cloud data,and the coarse registration transformation matrix is solved by singular value decomposition（ SVD）. The improved ICP algorithm is used to solve the final transform matrix. The experimental results show that the registration error is only 8.71895×10^-5 m^2,and the total time consuming is 6.61 seconds. Compared with the SIFT algorithm and the manual search matching point,the algorithm proposed is not only fast but also accurate.

关 键 词：三维人脸配准 SURF 奇异值分解 ICP 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]