Research on Reproducing Behaviors of Non-rigid Objects Present in Real Space in Virtual 3D Space

作为在虚拟3D空间中再现存在于真实的空间中的非刚性对象的行为的研究中的非刚性对象，如下所述研究（1）人图像和（2）弹性对象。(1)为了提高在放置在虚拟空间中的3D面部模型中再现其面部被摄像机观察的人的面部表情和面部运动的质量，一种方法从由上述摄像机获取的视频序列中的每一帧提取面部纹理，基于检测到的面部位置和姿势对提取的纹理应用仿射变换，然后将变换后的纹理映射到人脸模型上。其结果是，可以高质量地再现皱纹、眼睛和嘴等难以通过简单地使面部模型变形而真实地动画化的元素。 ...更多信息 .到目前为止，漫画肖像已经由漫画肖像作家绘制，但如果计算机可以自动生成漫画肖像，则可以期望实现逼真的面部复制。研究了最佳变形量与面部轮廓、眼睛和嘴部形状测量结果的关系，并从人体整体出发，研究了计算机视觉方法在真实的空间中识别人体运动的方法。更具体地，研究了用于识别未包括在用于识别处理的数据库中的未知人的运动的方法，以及用于识别与周围环境交互的人的动作的方法。(2)弹性对象为了实时稳定地表示三维虚拟弹性对象的大规模变形，提出了一种通过关节面连接多个弹性对象基元的关节式弹性对象模型。基于边界元法的方法可以稳定、实时地描述复杂形状弹性体的大范围变形。少

As the non-rigid objects in the research on reproducing behaviors of non-rigid objects present in real space in virtual 3D space, (1) human images, and (2) elastic objects are studied as described in the following.(1) Human ImagesTo improve the quality of reproducing the facial expressions and face movements of the person, whose face is observed by a camera, in the 3D face model placed in the virtual space, a method that extracts the facial texture from each frame in the video sequence acquired by the above-mentioned camera, applies Affine transform to the extracted texture based on the detected facial position and pose, and then maps the transformed texture onto the face model is proposed and studied. As a result, it turns out that the elements such as wrinkles, eyes and mouth, which are difficult to be animated realistically simply by deforming the face model, can be reproduced with high quality.Another methods that represent human faces realistically include the caricatured portrait … More . Till now, caricatured portraits have been drawn by caricature portrait writers, but if computers can generate caricatured portraits automatically, realistic reproductions of faces can be expected to be achieved. Relationship between optimal deformations and the results of the shape measurement of facial contour, eyes and mouth was studied.Concerning the entire human body, methods for recognizing human motions in real space by computer vision methods were studied. More specifically, a method for recognizing the motions of an unknown person, who is not included in the database for the recognition process, and a method for recognizing human actions interacting with the surroundings were researched.(2) Elastic objectsIn order to represent large-scale deformations of 3D virtual elastic objects in a stable manner in real-time, an articulated elastic object model that connects multiple elastic object primitives through joint surfaces is proposed. A method based on Boundary Element Method can represent large-scale deformations of elastic objects with complicated shapes in a stable manner in real-time. Less