%0 Journal Article
%T 基于单目RGB摄像机的三维人脸表情跟踪算法<br>3D facial expression tracking using a monocular RGB camera
%A 黄炎辉
%A 樊养余
%A 苏旭辉&lt
%A br&gt
%A HUANG Yanhui
%A FAN Yangyu
%A SU Xuhui
%J 山东大学学报（工学版）
%D 2017
%R 10.6040/j.issn.1672-3961.0.2016.466
%X 摘要： 利用单目摄像机对人脸表情进行三维跟踪有两个关键问题,一是从单幅图像中重建人脸的三维模型,二是建立不同三维模型之间稠密的对应关系。针对上述问题,提出一种有效的三维人脸表情跟踪算法。对输入的各帧人脸图像进行自动特征点检测,并根据2D特征点,利用Multilinear Model重建不同表情的三维模型。通过将三维模型重投影至图像空间,计算各帧图像间的Optic Flow与Sift Flow对应。根据图像间的对应关系,利用Laplacian网格变形对三维模型进行修正,建立不同表情之间三维的稠密对应。试验表明,该方法可以重建出较为真实的三维人脸表情模型,同时可以精确跟踪连续变化的三维表情序列。<br>Abstract: There were two key problems in tracking 3D facial expression using a monocular RGB camera. One was how to reconstruct 3D facial model from a single image, and the other was how to establish the dense correspondences between the different 3D models. To solve the above problems, an effective 3D facial expression tracking algorithm was proposed. The feature points were automatically detected on each input facial image frame, and the 3D models of different expressions were reconstructed by Multilinear Model according to 2D feature points. By re-projecting the 3D models into the image space, the Optic Flow and Sift Flow correspondences between images were calculated. According to the correspondences of the images, the 3D models were rectified by Laplacian mesh deformation, and the dense 3D correspondences between different expressions were established. The experimental results showed that the proposed method could create more realistic 3D expression models, and could accurately track the continuous change of the 3D expression sequence
%K 三维人脸表情跟踪
%K Laplacian网格变形
%K Multilinear Model
%K &lt
%K br&gt
%K 3D facial expressions tracking
%K Multilinear Model
%K Laplacian mesh deformation
%U http://gxbwk.njournal.sdu.edu.cn/CN/10.6040/j.issn.1672-3961.0.2016.466