CASIA OpenIR  > 毕业生  > 博士学位论文
基于视频图像的三维建模方法与系统----消化道三维建模方法与应用研究
其他题名3D Modeling Based on 2D Video Sequence ---- with Applications to Gastrointestinal Tract 3D Modeling Using Endoscopic Images
张鸿燕
学位类型工学博士
导师耿征
2011-05-25
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业控制理论与控制工程
关键词三维建模 消化道 无线胶囊内镜 流形建模 参数估计 图像拼接 三维重建 无线供电 3d Modeling Gastrointestinal Tract Wireless Capsule Endoscopy Manifold Modeling Parameter Estimation Image Mosaicing 3d Reconstruction Wireless Power Transfer
摘要基于视频图像的三维建模方法与系统是计算机视觉与医学图像处理中的重要研究内容。为了满足临床医学中对于消化道疾病诊断与分析的迫切要求, 本文对于人体消化道三维建模系统的一些基本问题做了比较深入系统的研究, 主要工作有: (1) 设计了消化道三维建模系统的架构。 (2) 提出了流形建模方法, 对其原理做了细致的分析, 给出了其在三维建模中的应用。事实上, 流形概念可以用摄像机投影模型来重新给予解释; 二维全景拼接算法与局部坐标配准均可用流形建模的语言进行统一描述。 (3) 给出了STLS 线性模型参数估计的算法实现与拓扑解释,考察了其与Levenberg-Marquardt 算法的相互关系; 结合分层运动模型与光流方程提出了彩色图像帧间单应变换矩阵的估计算法, 该方法在基于特征分析的估计算法失效时依旧可以得到良好的估计结果。 (4) 提出了适合于管道场景的二维全景图像拼接算法, 该方法既可用于消化道场景, 也适用于摄像机以轴向方式拍摄的其它场景。研究表明:管道图像的采集方式极大地影响了图像拼接算法的设计策略; 图像条带选取与整形是拼接算法中的必要步骤; 整形操作必需满足正交性约束与拓扑连通性变换两个要求; 复几何与Mobius 映射对于条带整形操作的实现极为重要。 (5) 以两视点度量重构为基础给出了三维度量重构中视点增加的递推算法; 提出了二维与三维处理相结合的建模方法, 兼顾了全景拼接与三维重建两者的优点, 更利于解决消化道疾病的诊断与治疗问题。 (6) 针对新型无线胶囊内镜系统的物理实现, 探索了成像光学系统与电能供应两类问题: 提出了环视全景镜头成像下的全景图像拼接算法与新的三维建模理论问题; 给出了消逝场无线供电的物理图像与性能分析。
其他摘要The study on 3D modeling based on 2D video sequence is of great significance in computer vision and biomedical image processing. In order to satisfy the pressing needs of gastrointestinal tract diseases diagnosing and treating in clinic medicine, the gastrointestinal tract 3D modeling system is studied systematically in this dissertation. The focus is put on some of the fundamental problems of 3D modeling approaches and systems, including scene manifold modeling, low-level processing algorithms, 2D and 3D image processing and physical implementation of new type wireless capsule endoscopy (WCE). The main works are summarized as following: (1) The architecture of the gastrointestinal tract 3D modeling system is established. (2) The manifold modeling approach and its application in 3D modeling are proposed and analyzed in detail for the first time. Actually, the concept of manifold in mathematics can be reinterpreted with camera projection in computer vision, and both the image mosaicing algorithm and local coordinate registration methods can be described in a unified way by manifold modeling. (3) The scaled total least squares (STLS) algorithm for linear model parameters estimation is implemented and its topological interpretation and relation with Levenberg-Marquardt algorithm are also presented. Furthermore, a novel estimation algorithm for intra-frame homography is put forward by combing the hierarchical motion model and optical flow equations, which works well when all the feature-based homography estimation methods fail. (4) The tubular scene manifold mosaicing (TSMM) algorithm and its revised version are proposed, which are applicable for the gastrointestinal tract scene and other scenarios in which the camera shots in an axial-way. The investigation shows that: the image acquisition procedure exerts a tremendous influence on the design strategies of image mosaicing; image strip selection and reshaping are two necessary steps in mosaicing algorithms; the orthogonality constraint and topological connectivity should be satisfied in image strip reshaping operation; and lastly, the complex geometry and Mobius mapping are valuable for the implementation of strip reshaping. (5) An iterative algorithm for adding views in metric 3D reconstruction is presented based on two-view 3D reconstruction. The key of the algorithm is the recursion formula for local coordinates alignment. The 2D+3D modeling method is proposed to combine the advantages of image mo...
馆藏号XWLW1656
其他标识符200718014628026
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6338
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
张鸿燕. 基于视频图像的三维建模方法与系统----消化道三维建模方法与应用研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2011.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20071801462802(4843KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[张鸿燕]的文章
百度学术
百度学术中相似的文章
[张鸿燕]的文章
必应学术
必应学术中相似的文章
[张鸿燕]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。