【作者】 李放；

【导师】 杨士强；

【作者基本信息】 清华大学 ， 计算机科学与技术， 2005， 硕士

【摘要】 多视点视频是近几年视频处理领域研究的热点方向。多视点视频是针对即将出现的交互式多媒体应用提出的,其所涵盖的双目立体视频与多视点视频播放将在未来几年中实用化,它将解决 3D 交互视频的表现、交互、存储和传输等问题。它的主要特点在于能够为用户提供交互选择观看视点的功能。虚拟视点的合成技术是提供交互功能的关键技术。虚拟视点的合成技术用来生成在视点切换过程中平滑过渡的虚拟图像。本文主要对视频内容的分层描述,视频对象的分割技术和虚拟视点图像的合成方法等内容进行了研究,并设计和实现了在大角度会聚拍摄条件的虚拟视点合成算法。论文的主要内容包括:(1)基于多视点视频的虚拟会议场景表示和前景提取与跟踪算法主要工作是建立半沉浸的投影显示设备构造虚拟环境,提出基于水线的视频对象分割算法,实现稳定的虚拟会议静态背景环境下与会者对象实时提取;采用线性视点合成算法,实现了基于位置的多视点视频重构合成。(2)多视点视频系统合成算法的设计与实现主要工作是在多视点视频系统之中合成算法的设计和实现。多视点视频系统中合成算法主要处理两部分内容:前景物体和背景。在合成的过程中采用不同的手段处理前景和背景。对于前景的合成算法主要分为三个步骤:图像特征的提取和跟踪,建立对应视频对象的不完全三维结构,生成中间图像的插值运算。对于背景的处理采用 sprite 的方法生成一幅摄像机系统覆盖场景的全景图。通过前后场景的融合形成虚拟视点图像。(3)多视点视频虚拟视点图像合成方法的改进改进的目的主要包括两个方面:合成图像的质量和合成算法的整体处理速度。合成虚拟图像的质量是最终评判算法的重要因素。我们主要考虑从图像特征跟踪的质量和数量以及图像中视频对象的边缘处理两个方面进行改进。关于算法的处理速度主要从系统级的模块设计,算法级的优化两个方面给出了设计方案。以上改进通过测试序列进行了验证。更多还原

【Abstract】 Multi-view video is a hot-point on the video processing research fieldrecently. It is proposed for interactive multimedia applications, includingbinocular three-dimensional video and multi-view video broadcasting whichwill come into being in the future. Multi-view video needs to solverepresentation, interaction, storage and transmission in 3D interactive video.Its feature focuses on the interaction function which is provided for the usersto select view point interactively. The synthesis of virtual view points is thekey technique in the multi-view video. It produces smooth virtual video in theprocess of view point transition. In this paper, we mainly focus on the layerrepresentation, video object segmentation, the virtual view point synthesis,and implementing the algorithm of the virtual view point synthesis in thecondition of large angel capture system.The main contributions of this thesis are as follows:(1) Research and implementation of the virtual teleconferencingrepresentation and foreground segmentation based on multi-view video.We construct virtual teleconferencing environment by usinghalf-immersion projection display device. We propose a watershed basedalgorithm which extracts stably real time video objects from the staticbackground. And then we implement position based multi-view videoreconstruction and synthesis.(2) Research and implementation of synthesis algorithm in multi-viewvideo system.The main task of this part is to design and implement the synthesisalgorithm in our multi-view video system. The algorithm deals with twokinds of data—the foreground data and the background data. For theforeground data, the algorithm consists of three parts: selecting and trackingfeature points;building incompletely 3D construction and interpolatingvirtual view point images. For the background data, we adopt sprite methodto produce a panoramic image. Finally we combine the two results to get thevirtual view point image.(3)Improvement of the synthesis algorithm in multi-view video systemThe aim of the improvement includes: increasing the quality of thesynthesis image and reducing the executing time of the whole algorithm. Inorder to increase the quality of the synthesis, we mainly improve the methodof tracking the feature points. In addition, to reduce the executing time, we domore improvements in both the system structure and the algorithm. And wetest it by using standard testing sequences.更多还原