基于视频序列的虚拟人渲染方法  

Rendering of virtual human based on video sequence

作　　者：淮鸽 刘越[1] 翁冬冬[1] 

机构地区：[1]北京理工大学光电学院北京市混合现实与新型显示工程技术研究中心,北京100081

出　　处：《光学技术》2015年第4期296-300,共5页Optical Technique

基　　金：国家自然科学基金(61370134);国家863计划项目(2013AA013901)

摘　　要：针对肢体动作有限的虚拟人仿真系统,提出了一种基于视频序列的虚拟人渲染方法。首先利用建模工具获取关键表情和口型的脸部模型,并对脸部模型进行预处理,在选取脸部模型后,采用基于模型的脸部姿态估计算法,将其叠加到包含肢体动作的图片上以生成具有新表情的图片;然后利用图像变形技术逐次生成含有虚拟表情的图像序列,并将图像序列转换为视频,通过播放视频渲染出逼真的虚拟人。实验结果表明,渲染得到的虚拟人的真实度有了明显的提高。To render the virtual human more realistically, a method based on video-sequence to get a virtual human is proposed. Every fragment of conversation is converted into a video according to the given transcript and the virtual human is rendered by playing the videos in order. Firstly, all the facial models are got by the 3D scanner and the models are clipped to get the mesh with smooth edge and the texture with transparent channel. As the models are ready, the position of the face is estimated in the picture and the selected models are embedded into the picture. Then, the transition between any adjacent frames is obtained by image morphing and all the image sequences are converted into a video. When playing the video, a virtual human like a real person can be seen to talk with us. Finally, the feasibility of the method is verified, and the effect of rendering is analyzed with a series of experiments, including the subjective evaluation and the calculation of matching between the real and composite pictures.

关 键 词：图像变形 脸部建模 虚拟化身 图像序列 

分 类 号：TN911.73[电子电信—通信与信息系统]