面向视频立体化应用的运动人体三维建模方法研究

项目名称： 面向视频立体化应用的运动人体三维建模方法研究

项目编号： No.61472362

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 王慧燕

作者单位： 浙江工商大学

项目金额： 83万元

中文摘要： 本项目的总体目标是面向单目视频立体化的应用需求，研究解决视频场景中多视图几何原理不适用的运动人体的深度恢复问题，以期在高效时空一致性视频分割技术、基于三维扫描的交互式人体三维形状和姿态建模以及服装建模、基于视频的人体三维运动恢复与几何模型重建、缺损或遮挡动态目标的时空一致性信息修复方法等几个方面的研究取得突破，获得视频序列中运动人体的精确深度信息，创建视频中着装运动人体的高质量三维模型，并集成课题组在自动摄像机跟踪与深度恢复、场景交互深度恢复等方面的研究成果，研发一个高效便捷的单目视频立体化原型系统，并探索其在电影、电视、广告、动漫等二维视频的立体化产业中的应用，促进计算机视觉与图形学的学科交叉与理论完善，为三维影视产业及相关产业的发展提供理论和技术支撑。

中文关键词： 三维立体化；三维重建；深度恢复

英文摘要： For the application demand of three-dimensional reconstruction of video, the main goals of this project are 1) to recover the depth information of dressed athletic human bodies showing in video which do not satisfy the general principle of multi-view epipolar constraint, in order to construct high quality 3D mesh models of athletic human body from monocular video series; 2) to assemble the achievements of our research team in the fields of automatic camera tracking and depth information recovering, interactive scene depth information recovering etc, in order to develop a prototype system of 3d modeling of video series. 3) to explore the application of 3d modeling techniques of movies, TVs, advertisements, and cartoons etc. To fulfill the goals, several problems from different research fields need be solved, such as, video segmentation techniques satisfying spatio-temporal information completion, interactive 3D human gesture and shape modeling and cloth modeling based on 3D scanning data, the human body movement recovering and geometric shape reconstruction based on video series, the spatio-temporal completion of defective dynamic scene based on video volume, etc.. The program can expected to promote the interdisciplinary fusion and theory development of computer vision and computer graphics, and provide the theoretic and technical support for the future development of three-dimensional film and television industry.

英文关键词： 2D to 3D conversion;3D reconstruction;depth recovery