视频三维立体化技术研究

项目名称： 视频三维立体化技术研究

项目编号： No.61272048

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 章国锋

作者单位： 浙江大学

项目金额： 83万元

中文摘要： 正在承担的青年科学基金项目主要是研究如何从图像/视频中恢复高质量的深度和三维信息，目前取得了重要进展，已在国际顶级期刊（TPAMI、TVCG）上发表论文2篇，视觉顶级会议（ICCV、CVPR、ECCV）上发表3篇，其它重要国际会议论文2篇（EI检索），国内核心期刊3篇，申请专利4项（其中1项已授权）。结合目前的视频三维立体化热潮，本项目将在原有的工作基础上继续深入和完善，研发一套完整的自动和交互相结合的视频三维立体化方法，解决视频的三维立体化难题。重点研究：复杂场景的时空一致性深度恢复技术，能够处理严重的遮挡、无特征区域以及动态场景的高质量深度恢复；快速的时空一致性视频交互分割与深度赋予技术；面向立体视频合成的时空一致性补全技术，解决基于图像的绘技术由于遮挡造成的图像内容缺失问题，保证生成的立体视频的高质量。在此基础上，本项目还将搭建起一个完整的视频三维立体化原型系统，验证所提出的方法。

中文关键词： 摄像机跟踪；深度恢复；三维重建；视频分割；立体视频

英文摘要： The Youth Science Foundation project mainly researches how to recover high-quality depth and 3D information from images/videos. We already have achieved important progress. We have published 2 papers in top international journals (TPAMI, TVCG), 3 papers in top vision conferences (ICCV, CVPR, ECCV), 2 papers in important international conferences (EI index), 3 papers in domestic core journals, and also applied 4 patents (one patent has been authorized). Considering the popularity of 3D videos, this project will continue and extend the original work, research and develop a series of automatic and semi-automatic 3D conversion methods to solve the 3D video conversion problem. First, we would like to develop a spatio-temporally consistent depth recovery method which can handle complex scenes, such as strong occlusions, large textureless regions and dynamic scenes. Second, we would like to develop a fast interactive video segmentation and depth recovery method, by combining vision, interactive graphics and machine learning algorithms. Third, we would like to develop a spatio-temporal video completion method to complete the missing pixels caused by occlusions during novel view synthesis for generating stereoscopic images, so that a high-quality stereoscopic video can be achieved. Finally, we shall build a complete 2D t

英文关键词： camera tracking；depth recovery；3D reconstruction；video segmentation；stereoscopic video