完整虚拟人头部建模及面部表情合成研究

发布时间：2018-04-14 08:53

本文选题：完整头部网格模型 + 模型调整　；参考：《西南交通大学》2016年硕士论文

【摘要】：随着虚拟现实技术、计算机图形学技术和计算机视觉技术的快速发展,三维虚拟人逐渐发展成为一门新兴的学科,并吸引了越来越多的人从事相关的研究。由于人类的面部表情和情感非常丰富,所以三维虚拟人头部的研究更是成为了研究的热点。三维虚拟人头部的研究不仅具有巨大的研究价值,还具有广阔的应用前景,比如在医学整容、电子商务、视频教学、游戏开发等领域总能看到它的身影。在研究中,人们常常利用三维扫描仪或者建模软件来建立头部模型,但是这些方法建立的模型都存在一些问题,最主要的就是数据量太大,不容易控制。因此,建立一个数据量比较少且能够完整地表达头部的结构,同时又易于控制的头部模型就成为本文的研究重点。随着研究人员的增加和相关工作的积累,出现了很多不同的头部模型,这些模型要么是结构不够完善,要么就是模型过于复杂,不容易控制。针对该问题,本文将建立一个结构完整能表达头部特征的模型。本文首先将软件导出模型简化并去除面部网格后的模型和标准Candide-3模型结合在一起,合成一个通用的模型。然后,利用Bezier曲线模拟眼睑的内外边缘和上眼睑沟实现眼睑的重建,同时利用两个半径不同的球体相结合的方法完成眼球模型的建立,并用四角面片模拟牙齿,就可以得到一个完整的虚拟人头部模型。接着,利用从输入的单幅正面人脸照片上提取的特征点信息对模型面部的特征点、面部非特征点以及后脑勺的点进行调整,得到一个和输入照片的人脸相匹配的特定头部模型,利用不同的方法对不同的模块贴纹理,就可以得到一个真实的头部模型。最后,利用人脸动作单元对头部模型面部和表情合成相关的顶点坐标进行调整,合成面部真实感的表情动画。实验结果表明,建立的头部模型能够给人很强的真实感,且模型容易控制,基于该模型合成的虚拟人面部动画比较真实。
[Abstract]:With the rapid development of virtual reality technology, computer graphics technology and computer vision technology, 3D virtual human has gradually developed into a new subject and attracted more and more people to engage in related research.Because human facial expressions and emotions are very rich, the research of 3D virtual human head has become a hot topic.The research of 3D virtual human head not only has great research value, but also has a broad application prospect, such as medical cosmetic surgery, e-commerce, video teaching, game development and other fields can always see its figure.In the research, people often use 3D scanner or modeling software to build the head model, but there are some problems in these methods, the most important thing is that the data is too large to be easily controlled.Therefore, it is the focus of this paper to establish a head model which has less data volume and can express the head structure completely and is easy to control at the same time.With the increase of researchers and the accumulation of related work, many different head models have emerged. These models are either imperfect in structure or too complex to be easily controlled.In order to solve this problem, this paper will establish a structural integrity model which can express the features of the head.In this paper, the software derived model is simplified, and the model after removing the facial mesh is combined with the standard Candide-3 model to synthesize a general model.Then Bezier curve was used to simulate the inner and outer edges of the eyelid and the upper eyelid sulcus to reconstruct the eyelid. At the same time, the eye model was established by combining two spheres with different radii, and the teeth were simulated by the quadrangular slices.A complete virtual human head model can be obtained.Then, the feature points of the face, the non-feature points of the face and the points in the back of the head are adjusted using the feature points information extracted from the input single frontal face image, and a specific head model matching the face of the input photo is obtained.Using different methods to paste texture to different modules, we can get a real head model.Finally, the facial action unit is used to adjust the vertex coordinates of facial and facial expression synthesis to synthesize facial realistic facial animation.The experimental results show that the established head model can give a strong sense of reality, and the model is easy to control, and the virtual human facial animation based on the model is more realistic.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】