一种无标记点人脸表情捕捉与重现算法  

A Markerless Facial Expression Capture and Reproduce Algorithm

作　　者：吴晓军[1,2] 鞠光亮 

机构地区：[1]哈尔滨工业大学深圳研究生院,广东深圳518055 [2]深圳先进运动控制技术与现代自动化装备重点实验室,广东深圳518055

出　　处：《电子学报》2016年第9期2141-2147,共7页Acta Electronica Sinica

基　　金：深圳市科技计划项目(No.JCYJ20150928162432701;No.JCYJ20140417172417166)

摘　　要：提出了一种无标记点的人脸表情捕捉方法.首先根据ASM(Active Shape Model)人脸特征点生成了覆盖人脸85%面部特征的人脸均匀网格模型;其次,基于此人脸模型提出了一种表情捕捉方法,使用光流跟踪特征点的位移变化并辅以粒子滤波稳定其跟踪结果,以特征点的位移变化驱动网格整体变化,作为网格跟踪的初始值,使用网格的形变算法作为网格的驱动方式.最后,以捕捉到的表情变化数据驱动不同的人脸模型,根据模型的维数不同使用不同的驱动方法来实现表情动画重现,实验结果表明,提出的算法能很好地捕捉人脸表情,将捕捉到的表情映射到二维卡通人脸和三维虚拟人脸模型都能取得较好的动画效果.This paper presents a markerless facial expression capture and reproduce algorithm. Firstly,an uniform mesh model is built based on the feature points from ASM（ Active Shape Model）. It can cover 85 percent of the face. Then,a method to capture facial expression based on the face model is proposed. The optical flowtracking is used to track the feature points with particle filter for stablizing the result. The feature points＇ displacement can drive the overall mesh to deform as the initial value of the mesh tracking. Finally,the captured expression data is used to drive face models with different methods for the models of different dimensions,and then the facial animation can be reconstructed. Experimental results showthat the proposed algorithm can capture facial expressions well,and the animation effect is good when mapping the captured expression to 2D cartoon or 3D virtual face models.

关 键 词：无标记点跟踪 人脸表情捕捉 表情动画重现 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]