面向普通用户的3D虚拟人脸动画  

3D Virtual Facial Animation for General Users

作　　者：罗常伟[1] 江辰[1,2] 李睿[1,2] 於俊[1,3] 汪增福[1,2,3] 

机构地区：[1]中国科学技术大学自动化系,合肥230027 [2]中国科学院合肥智能机械研究所仿生感知与控制研究中心,合肥230031 [3]中国科学技术大学语音及语言信息处理国家工程实验室,合肥230027

出　　处：《计算机辅助设计与图形学学报》2015年第3期492-498,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61472393);安徽省自主创新专项资金智能语音技术研发和产业化专项(13Z02008-5)

摘　　要：为合成真实感人脸动画,提出一种实时的基于单摄像头的3D虚拟人脸动画方法.首先根据用户的单张正面人脸图像重建用户的3D人脸模型,并基于该人脸模型合成姿态、光照和表情变化的人脸图像,利用这些图像训练特定用户的局部纹理模型;然后使用摄像头拍摄人脸视频,利用特定用户的局部纹理模型跟踪人脸特征点;最后由跟踪结果和3D关键形状估计Blendshape系数,通过Blendshape模型合成的人脸动画.实验结果表明,该方法能实时合成真实感3D人脸动画,且只需要一个普通的摄像头,非常适合普通用户使用.To realistically animate a 3D virtual face, we propose a real-time video-driven facial animation method. Firstly, we reconstruct the user＇s 3D face model from a single frontal face image, and synthesize a set of facial images with different pose, illumination and expression based on this face model. The synthetic images are used to train a user-specific local texture model. Then, a video camera is used to record the facial performance of the user. Facial feature points are tracked using a constrained local model based method. Finally, facial tracking results are transformed to Blendshape coefficients based on 3D key shapes, and facial animation is generated by using Blendshape interpolation. Experiments show that our method is effective for realistic facial animation synthesis. Moreover, the method only requires a web camera and can be easily used by an ordinary user.

关 键 词：人脸动画 人脸特征点跟踪 Blendshape模型 表演驱动 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]