权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

HIGH FIDELITY FACIAL MUSCLE MODELING AND ITS DYNAMIC CONTROL BY ANATOMICALAPPROACH

高保真面部肌肉建模及其解剖学动态控制

基本信息

批准号：
13450161
负责人：
MORISHIMA Shigeo
金额：
$ 7.55万
依托单位：
Waseda University (2004)Seikei University (2001-2003)
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (B)
财政年份：
2001
资助国家：
日本
起止时间：
2001 至 2004
项目状态：
已结题

项目摘要

The purpose of this research is to realize a high quality facial animation synthesis by constructing personal facial muscle model which simulate high fidelity muscle tissues' dynamics. Each flat facial muscle tissue is modeled by group of springs spread on a plane and the motion of each muscle is simulated by solving motion equation by considering force and contraction. Collision detection of each muscle, resistance force from skull surface and muscle volume are also considered to generate natural model of facial dynamics.Objective evaluation method of muscle action using EMG is proposed and also evaluation method of lip synchronization using noise added voice is proposed.To customize a generic facial muscle model to each person, GUI based face fitting tool is designed. Firstly by putting feature points to captured face texture manually, automatically generic facial muscle model is modified to fit onto personal range data. Then 3D location of each muscle tissue is also customized. By minor adjustment of each muscle location and checking motion of face skin surface, personal facial muscle model is completed.Facial animation control tool by manipulating each muscle is constructed to generate several variation of face expression. And we define each basic target face considering FACS by combination of facial muscle spring forces. This result is included in the face synthesis module in the open source toollkit for anthropomorphic spoken dialog agent. And some of the result from this research contributed to "Future Cast System" which is developed for Expo 2005.

本研究的目的是通过构造逼真的肌肉组织动态模型，实现高质量的人脸动画合成。每个扁平的面部肌肉组织由一组在平面上伸展的弹簧来模拟，并且通过考虑力和收缩来求解运动方程来模拟每个肌肉的运动。在此基础上，提出了基于肌电信号的肌肉动作客观评价方法和基于噪声语音的唇同步评价方法，并设计了基于GUI的人脸拟合工具，为每个人定制一个通用的面部肌肉模型。该方法首先通过人工对采集到的人脸纹理进行特征点匹配，自动对通用的人脸肌肉模型进行修改，使其与个人范围数据相匹配。然后还定制每个肌肉组织的3D位置。通过对各个肌肉位置的微调和对人脸皮肤表面运动的检测，完成了个性化的人脸肌肉模型，并通过对各个肌肉的操纵构造了人脸动画控制工具，生成多种表情变化。并结合面部肌肉弹性力定义了基于FACS的基本目标人脸。这个结果包含在开放源代码工具包中的人脸合成模块中，用于拟人口语对话代理。本研究的部分成果为2005年世博会开发的“未来投景系统”做出了贡献。