基于视频的三维重建研究
这是一篇华中科技大学的硕士毕业论文,里面对三维重建技术的讲解很详细,适合作为综述来看独创性声明本人声明所呈交的学位论文是我个人在导师指导下进行的研究工作及取得的研究成果。尽我所知,除文中已经标明引用的内容外,本论文不包含任何其他个人或集休已经发表或撰写过的研究成果。对木文的研究做出贡献的个人和集休,均已在文中以明确方式标明。本人完全意识到本声明的法律结果由本人承担。学位论文作者签名:日期:年月日学位论文版权使用授权书本学位论文作者完全了解学校有关侏留、使用学位论文的规定,即:学校有权保留并向国家有关部门或机构送交论文的复印件和电子版,允许论文被查阅和借阅。本人授权华屮科技大学可以将本学位论文的全部或部分内容编入有关数据库进行检索,可以釆用影印、缩印或扫描等复制手段保存和汇编本学位论文。保密口,在年解密后适用本授权书本论文属于不保密口。(请在以上方框内打“√”)学位论文作者签名指导教师签名:日期:年月日日期:年月日万方数据华中科技大学硕士学位论文摘要基于视频的重建技术在计算杋视觉领域中扮演着非常重要的角色,而如何恢复场景的三维模型是目前研究的热点与难点问题。本文围绕基于视频的场景重建技术展开讨论,包括棊于单目视频的三维场景重建和于双目视频的视差图和场景流获取。由于单目包含的深度信息比较少,如何基于单目视频恢复相机的运动参数以及目标的深度信息是研究的重与难点。双目视频虽然包含了非常显著的深度信息,但是考虑到视频中场景的迕续性问题,如何使得恢复岀的深度图保持前后帧的连续性以及场景中运动日标的一致性,也是比较困难的问题。因此,针对上述所提到的问题进行了深入的研究,具体的研究工作如下第一,对三维重建研究进行了详细的介绍,介绍了对于特征点匹配的理解以及我们提出的基于特征引导偏向性高斯混合模型( Feature Guided Biased GaussianMixture model,FGBG);详细介绍立体视觉中立体匹配算法的原理、分类及评测标准,并在4个典型的数据集上对有代表性的局部、全局、半全局算法进行对比实验。此外,详细介绍运动恢复结构(SFM)的基本原理,并进行了实验分析。第二,提岀一种基于双目视频的视差图和场景流获取技术。基于双目视频,首先获得初始的视差图和2D特征点轨迹;在此基础上获得初始的3D稀疏运动轨迹,利用本文提出的 Object Motion Hypothesis(OMH)算法获得运动物体的致性假设采用 slanted-plane model以及参考图像与前后时间点图像对的约束关系,构建超像素和运动物体之间的能量模型,通过优化获得视差和场景流的估计结果。第三,提出一种棊于单目视频的动态场景重建系统。在获取特征点轨迹的基础上,基于运动信息获得特征点轨迹的聚类结果;提出一种基于超像素的多标记Graph-cut算法,得到每一个日标的精确边界;为每一个运动日标分配一个虚拟相机通过标准的SFM方法分别单独估计每个运动目标对应的虚拟相机的参数和稀疏三维点云,通过PMVS和泊松表面重建获得目标的稠密重建结果。关键词:三维重建、单∏视频、双目视频、视差、场景流万方数据华中科技大学硕士学位论文Abstract3D reconstruction based on video has play an important role in computer vision, andhow to recover 3D scene model has been paid much attention and is a difficult problemBased on the importance of 3D reconstruction, in this paper, the 3D reconstruction basedon video has been studied, including 3D scene reconstruction based on monocular videoand depth map and scene flow estimating based on binocular video. Since the monocularcontains much less depth information, how to recover the camera motion and depth maphas been a difficult problem. Besides, although binocular view contains significant depthinformation, it is difficult to keep the consistency of depth map and moving objectsTherefore, in view of the problems mentioned above the specific research works are asFirst. we introduce two directions of 3D reconstruction in computer vision: based onstereo vision method and based on structure from motion. The stereo matching method hasbcen introduced in detail, including algorithm principle, classification, and evaluationmethod. And, we compare the global, local and semi-global algorithm on four typicaldataset. In addition, we have made a detail introduction of structure from motion(SFM)and the experiment has been carried out to get 3D point cloudSecond, a method for depth map and scene flow estimation is proposed. First, inputbinocular video, initial disparity map is got by SGM, 2 point trajectories are got byoptical flow. Then the 3D tracks are got by disparity map and 2D point trajectories, get theobject motion hypothesis. Considering constraint between the reference image and theforward-backward images, the energy model based on super-pixel and object isconstructed using slanted plane model. Finally, the depth map and scene flow will be gotThird, a method for reconstructing monocular dynamic scene with multiple movingrigid objects captured by a single moving camera is proposed. First of all, feature pointsare matched through the video sequence via the optical flow method and the tracks "aregot based on these matches. Then the tracks are divided into several groups according totheir motion differences. An improved graph cuts based multi-label auto imagesegmentation method is used to acquire the accurate boundary of each moving object and万方数据华中科技大学硕士学位论文the static background. Then we assume a virtual camera for each moving object and thestatic background. The pose of these virtual cameras are estimated via the standardStructure from Motion(SFM) pipeline. Finally a dense point set and textured model isreturned for each virtual camera. We evaluate our approach on real-world video sequenceand demonstrate its robustness and effectivenessKey words: 3D reconstruction, monocular video, binocular video, disparity, scenefleOw万方数据华中科技大学硕士学位论文目录摘要Abstract绪论1研究的背景及意义2国内外研究现状1.3论文的主要工作及结构···································:··········.················4·2三维重建基本方法研究2.1引言.………8)2.,2线性摄像机模型(8)23基于特征点的图像匹配24运动恢复结构方法(12)2.5立体匹配与三维重建···.·.·······.·················:····.····················(15)26本章小结(22)基于双目视频的视差图与场景流估计3.1引言(23)3.2运动目标的提取(25)3.3双向约束场景流模型..31)34实验分析.333.5本章小结(444基于单目视频的三维重建研究(45)4.2目标分割(464.3三维场景估计(51)万方数据华中科技大学硕士学位论文4.4实验分析(52)4.5本章小结(55)5全文总结与展望5.1木文的主要页献与创新点(56)5.2工作展望…7)致谢S8)参考文献非D·非非··非。非(59)附录万方数据华中科技大学硕士学位论文绪论11研究的背景及意义视觉是人类的基本功能。通过视觉,人们能够感知外部世界中物体的大小,以及辨别物体之间的相对位置,并且了解它们之间的相互关系。人类把这种功能称为视觉功能。随着科学技术的不断创新,新兴的电子产品不断涌现,数码设备的成熟和计算机理论的涌现让人们越来越关注计算机视觉。人们开始利用摄像机采集视频或者图像,并将其转化为人类可理解的信号。即利用计算机实现模仿人类视觉的功能,计算机视觉也就随之六生。计算机视觉是个涵盖多种学科知识的新兴学科。其理论研究的最终目的是通过对采集到的视频或者图像进行处理,将二维图像或视频转化为三维信息,从而感知场景或物体的形状及运动。因此,计算机视觉吸引了越来越多的研究人员参与其中,包括图像处理与模式识别,应用数学,计算札科学与技术等等。三维场景重建作为计算札视觉中一个重要的研究方向,受到许多研究者的青睐。最近,获取三维场景信息的方式主要有以下三种:第一种,利川常见的建模软件3DMax、CAD等进行重建;第二种,利用深度扫描仪、红外或者激光测距仪器等设备进行三维重建;第三种,利用计算机视觉原理,基于视频或者图像获取场景的三位模型。在上述方法中,第一种是最为成熟的,但是第一种方法的操作步骤十分复杂,并且建模周期长。第二种方式能够获得物伓的髙精度几何模型,但是这些仪器价格昂贵,费时费力,并且对于重建大型场景非常局限。因此,第三种方式受到了普遍的关注,它可以重建复杂的室外大型场景,真实感强,价格低廉且方便携带。利用图像或者视频对场景进行重建,即从图像或视频中恢复场景或者物体的三维几何信息,构建三维模型,给人以视觉亨受。三维重建的用途十分广泛,它可以用于机器人导航,无人驾驶,医学图像分析,游戏等众多方向在众多的三维场景重建方法中,于视频的重建方法一直是一个研究热点。其中,从单目视觉的角度出发,基于单目视频的三维重建技术就是利用单个摄像札对万方数据华中科技大学硕士学位论文场景进行拍摄,研宄如何利用图像序列光流估计运动物体或场景的三维运动来重建三维模型。从双日视觉的角度出发,基于双∏视频的三维重建技术就是利用两个摄像机,从两个不同的角度对同一个场景进行拍摄,研究如何利用左右两个图像序列各自的运动信息,以及左右视图之间的视差信息,完成场景的三维重建。本文的基于视频的三维重建技术具有十分重要的研究价值。针对双目视频,提出了一种基于双目视频的视差图和场景流获取技术,目的是同时获得视差图和场景流信息、。针对单目视频,提出个完整的基于包含多个刚体运动目标的单目动态场景视频的重建系统。12国内外研究现状121基于单目视觉的三维重建研究现状近年来,3D静态场景的重建己经取得了显著性的突破。其中,大多数的研究都是遵循一个特定的步骤:首先从一组多视角的图像中提取特征点,然后对多视图中的特征点进行匹配,构建基础矩阵,恢复相机参数,从而得到玚景的三维结构凹。其中, Snavely N主要通过SFM( (structure from motion)从无序图像序列中恢复相机的位置以及获得场景的三维稀疏点云倒。除∫稀疏点云的重建之外,很多学者也集中研究场景的三维稠密重建四。其中, Seitz s m对多种立体匹配算法进行比较,并且是第一个提供已标定的多视图数据集。 Kolev K在前者的基础之上提出了一个全局能量模型,融合了轮廪信息和立体信息。值得一提的是,深度信息也是一种非常有前景的3D重建方法,主要思想是通过恢复图像的深度信息,融合多幅深度图逃行稠密重建η。此外,很多研究集屮于基于单个视频的稠密表面重建,主要包括基于场景流( scene flow)s, mesh- based稠密表面重建例, patch-base稠密表面重。但是,大多数捕获的视频中,动态场景视频比铰常见。而上述的研究只能用于处理静态场景,它们在应对多目标运动场景方面是十分有限的。最近, Tron r提出了一个包含动态运动目标的场景分割标准山,它是·个重要的3D运动估计和重建的预处埋过程视频重建主要有于两个视图12和基于多个视图314其中,HanM和万方数据
- 2020-12-11下载
- 积分:1
Key Technologies for 5G Wireless Systems
5G无线通信系统关键技术(剑桥大学出版社) 2017年出版 对于5G所有最新技术进行了详细说明 很全的工具书Key Technologies for5G Wireless SystemsVINCENT W. S, WONGUniversity of British ColumbiaROBERT SCHOBERUniversity of Erlangen-NurembergDERRICK WING KWAN NGUniversity of New South WalesLI-CHUN WANGNational Chiao-Tung University即CAMBRIDGEUNIVERSITY PRESSCAMBRIDGEUNIVERSITY PRESSUniversity Printing House. Cambridge CB2 SBS. United KindomOne Liberty Plaza, 20h Floor New York, NY I(H0X, USA477 williamstown Road, port Melbourne, yic 3207 australia48424, 2nd Floor, Ansar Rod, Daryaganj. Delhi- I l4XH2, India79 Anson Road, #o6-(/ 00, Singapore 079%MCambridge University Press is part of the Lniversity of CambridgeIt furthers the University s mission by disseminating knowledge in the pursuit ofeducation, leaming and research at the highest international levels of excellence.www.cermbrid吧eInformtiononthistitlewww.cambridgeorg/978110713241810,1017③781316771655C Cambridge University Press 2017This puhlication is in copyright. Subjcct to sututonry exceptionand to the provisions of relewant collective licensing agreementsno reproduction of any part may take place without the writtenpermission of Cutmbridgre University Press.First published 2(117Printed in the United Kingdom by TJ International Ltd. Padstow, CornwallA catalogue recor for this pudlieafiove is aailable fromm the British LibraryLibrary of Congress Cataloging- in Pi hlicaiomz dataNames: Wong, Vincent W.S., editorTitle: Key technologies for 5G wireless systems/edited by Vincent W.S. Wong [and 3 otherOther titles key technologies for five g wireless svstemsDescription: Carmbrisige: New York, NY: Cambridge Lniversity Press, 2017.Identifiers: l CCN 2016045220)1 ISBN 9781 172418 (hardback)Subjects: LCSH: Wireless communication systems, I Machine-to-machinecommunications. Internet of things.Classitication: LCC TKs1032K49 2(17 DDC 621.38450-dc23LcrecordavailaBleathttps://lccnioc-gov/2016m5220)ISBN 978-1-107-17241- HardbackCambridge University Press has no responsibility for the persistence or accuracy ofURLs for extermal or third-party Internet websites referred to in this puhlication,and does not guarantee that any content on such websites is, or will remainaccurate of appropriateContentsList of Contributorspage xvIPrefaceKXIOverview of New Technolog ies for 5G SystemsVincent W S, Wong, Robert Schober, Derrick Wing Kwan Ng, and Li-Chun Wang1.1 Introduction1.2 Cloud Radio Access Networks1.3 Cloud Computing and Fog Computing1. 4 Non-orthogonal Multiple Access1. 5 Flexible Physical Layer Design334.4671. 6 Massive MIMo1. 7 Full-Duplex Communications1. 8 Millimeter wave1.9 Mobile Data Offloading, LTE-Unlicensed, and Smart Data Pricing131. 10 IoT M2M. and D2D1. I1 Radio Resource Management, Interference Mitigation, and Caching61. 12 Energy Harvesting Communications1. 13 Visible Light Communication19Acknowledgments20ReferencesPart I Communication Network Architectures for 5G Systems25Cloud Radio Access Networks for 5G Systems27Chih-Lin I, Jinn Huang, Xueyan Husang, Rongwved Ren, and Yami. Chen2.1 Rethinking the Fundamentals for 5G Systems272 User- Centric Networks2923 C-RAN Basics292.3.1 C-RAN Challenges Toward SGI302.4 Next Generation Fronthaul Interface (NGFI: The FH Solutionfor SGC-RAN312. 4.1 Proof-of-Concept Development of NGFI33Contents2.5 Proof-of-Concept Verification of Virtualized C-RAN2.5.1 Data packets3725.2 Test Procedure382.5.3 Test Results392. 6 Rethinking the Protocol Stack for C-RAN2.6.1 Motivation402.6.2 Multilevel Centralized and Distributed Protocol Stack402.7 Conclusion45AcknowledgmentsReferencesFronthaul-Aware Design for Cloud Radio Access Networks48Liang Liu, Wei Yu, and Osvaldo Simeone3. 1 Introduction483.2 Fronthaul-Aware Cooperative Transmission and Reception493. 2.1 Uplink513.2.2 Downlink573.3 Fronthaul-Aware Data Link and Physical layers61.3. I Uplink633.3.2 Downlink693.4 Conclusion73Acknowledgments74References74MobEdge computing76Ben Liang4.1 Introduction764.2 Mobile Edge Computing774.3 Reference architecture794.4 Benefits and Application Scenarios804 4.1 User-Oriented Use cases4. 4.2 Operator-Oriented Use Ca814 5 Research challenges824.5.1 Computation Offloading824.5.2 Communication Access to Computational Resources834.5.3 Multi-resource Schedulin844.5 4 Mobility Management854.5.5 Resource Allocation and Pricing4.5.6 Network functions virtualization864.5, 7 Security and Pri864.5.8 Integration with Emerging Technologies874.6 Conclusion88ReferencesContentsDecentralized Radio Resource Management for Dense HeterogeneousWireless networksAbolfazl Mehhodniya and Fumiyuki Adach5.1 Introduction925.2 System Model935.2.1 SINR Expression5.2.2 Load and Cost Function Expressions955.3 Joint BSCSA/UECSA ON/OFF Switching Scheme965.3.1 StrateTy Selection and Beacon Transmission53.2 UE AssocIation5.3.3 Proposed Channel Segregation Algorithms985.3.4 Mixed-Strategy Update3.4 Computer Simulation5.5 Conclusion104Acknowledgments04References105Part ll Physical Layer Communication Techniques107Non-Orthogonal Multiple Access(NOMA)for 5G Systems109Wei Llang, Zhiguo Ding, and H. Vincent Poor6.1 Introduction1106.2 NOMA in Single-Input Single-Output(SISO)Systems1126.2.1 The basics of nomaI126. 2. 2 Impact of User Pairing on NOMA136.2,3 Cognitive Radio Inspired NOMA6. 3 NOMA in MIMO Systems1206.3.1 System Model for MIMO-NOMA Schemes1216.3.2 Design of Precoding and Detection Matrices with Limited CSIT 1236.3.3 Design of Precoding and Detection Matrices with Perfect CSIT 1266.4 Summary and Future Directions128ReferencesFlexible Physical Layer Design133Maximilian Matthe, Martin Danneberg, Dan Zhang, and Gerhard Fettweis7.1 Introduction1337. 2 Generalized Frequency Division Multiplexing357.3 Software-Defined waveform1377. 3. 1 Time Domain Processing1387.3.2 Implementation Architecture1387.4 GFDM Receiver Design14174 Synchronization unit1427. 4.2 Channel Estimation Unit1474.3 MIMo-GFDM Detection Unit145Contents7.5 Summary and Outlook147Acknowledgments148References488Distributed Massive MIMO in Cellular Networks15IMichail Matthaiou and Shi Jin8. I Introduction15l8. 2 Massive MIMO: Basic Principles1528.2.1 Uplink Downlink Channel Models1538.2.2Favorable Propagation1548.3 Performance of Linear Receivers in a Massive MIMO Uplink1548.4 performance of linear precoders in a massive mimo downlink1578. s Channel estimation in massive mimo systems1588.5.1 Uplink Transmission1598.5.2 Downlink Transmission1608.6 Applications of Massive MIMO Technology1618.6.1 Full-Duplex Relaying with Massive Antenna Arrays1618.6.2 Joint Wireless Information Transfer and Energy Transfer forDistributed massive mimo1638.7 Open Future Research Directions1678. 8 Conclusionl68References169Full-Duplex Protocol Design for 5G Networks172Tanelf Ahonen and Risto wichman9.1 Introduction1729. 2 Basics of Full-Duplex Systems1739.2.1 In-Band Full-Duplex Operation Mode1739.2.2 Self-Interference and Co-channel Interference1749.2.3 Full-Duplex Transceivers in Communication Links1759. 2. 4 Other Applications of Full-Duplex Transceivers1789.3 Design of Full-Duplex Protocols1799.3, 1 Challenges and Opportunities in Full-Duplex Operation1799.3.2 Full-Duplex Communication Scenarios in 5G NetworksR9.4 Analysis of Full-Duplex Protocols1829.4.1 Operation Modes in Wideband Fading Channels1829. 4, 2 Full- Duplex Versus Half-Duplex in Wideband Transmission1849.5 Conclusion1849.5.1 Prospective Scientific Research DirectionsI849.5.2 Full-Duplex in Commercial 5G Networks185RLItrtncekl8610Millimeter Wave Communications for 5G Networks188Jiho Song, Miguel R Castellanos, and David J. LoweContentsⅸx10.1 Motivations and Opportunities18810.2 Millimeter Wave Radio Propagation18910. 2.1 Radio Attenuation1890. 2. 2. Free-Space Path LOSs19I10.2.3 Severe shadow19310.2 4 Millimeter Wave Channel model19310.2.5 Link Budget Analysis19410.3 Beamforming Architectures19510.3, Analog beamforming solutions19610.3.2 Hybrid Beamforming Solutions20010.3.3 Low-Resolution Receiver Architecture2010.4 Channel Acquisition Techniques20110.4.1 Subspace Sampling for Beam Alignment20210.4.2 Compressed Channel estimation Techniques20510.5 Deployment Challenges and Applications20710.5.1 EM Exposure at Millimeter Wave Frequencies20710.5.2 Heterogeneous and Small-Cell Networks208Acknowledgments209References209Interference Mitigation Techniques for Wireless Networks214Koralia N Pappi and George K, Karag annidis1 1.1 Introduction21411.2 The Interference Management Challenge in the 5G vision21411. 2. 1 The 5G Primary Goals and Their Impact on Interference2141 1.2.2 Enabling Technologies for Improving Network Efficiencyand Mitigating Interference21611.3 Improving the Cell-Edge User Experience: Coordinated Multipoint218I 1.3.1 Deployment Scenarios and Network Architecture2181 13. 2 CoMP Techniques for the Uplink22011.3.3 CoMP Techniques for the Downlink2211 1.4 Interference Alignment: Exploiting Signal Space Dimensions2231 1.4.1 The Concept of Linear Interference Alignment224L1. 4.2 The Example of the X-Channel225I 1. 4.3 The K-User Interference Channel and Cellular NetworksAsymptotic Interference Alignment22611.4.4 Cooperative Interferenee Networks22711.4.5 Insight from IA into the Capacity Limits of Wireless Networks 22711.5 Compute-and-Forward Protocol: Cooperation at the ReceiverSide for the Uplink22811.5.1 Encoding and Decoding of the CoF Protocol22811.5.2 Achievable-Rate Region and Integer Equation Selection23011.5.3 Advantages and Challenges of the CoF Protocol232IL6 Conclusion233References233
- 2020-12-06下载
- 积分:1