节点文献

基于GPU编程的虚拟自然环境技术研究

Virtual Natural Environment Technique Research Based on GPU Programme

【作者】 韩元利

【导师】 胡鹏;

【作者基本信息】 武汉大学 , 地图学与地理信息系统, 2007, 博士

【摘要】 GPU以超越半导体技术摩尔定律三倍的速度快速发展,为开发实现大型实时三维仿真系统提供了一个前所未有的新手段,技术的提高也直接带动了人们对三维仿真系统认知的提升。虚拟自然环境超越了VR对空间实体的表达,包含了虚拟地理环境、虚拟物理环境、虚拟气象环境等众多抽象但客观存在的主题,提出了更具沉浸感、更全面的仿真空间表达环境。本文以综合当前国内外先进的技术为基础,结合先进的GPU编程方法实现,目的在于探讨建立一套完整的虚拟自然环境模拟实现的技术体系,研究的内容主要集中在虚拟地理环境、虚拟物理环境和虚拟气象环境的模型表达与GPU编程技术实现。本文在组织上各章节基本独立,主题分明,但又紧紧围绕虚拟自然环境的研究内容与实现手段进行组织。在本文的第一章中,我们论述了空间认知观的发展过程,并认为技术发展与手段更新是影响空间信息表达的关键因素;基于这样的认识,结合当前先进的GPU图形开发系统强大的性能提升,我们提出了"虚拟世界"的认知体系结构,将虚拟世界分为可感知的“虚拟自然环境”与“虚拟社会”两个体系,并且指出它们的分割标准是以“是否有人参与”为原则进行的。作为本文的主要研究内容,我们分析了虚拟自然环境的组成体系,初步确立了虚拟地理环境、虚拟物理环境与虚拟气象环境三个主题作为虚拟自然环境的研究内容,论述了各个主题的研究内容与实现技术路线。针对“虚拟现实”与“虚拟世界”的概念,我们辩证地论述了虚拟现实技术与虚拟世界(包括其各个研究分支)的关系问题,明晰了"虚拟世界"是研究目标,而虚拟现实技术只是一个研究手段、一种实现途径;作为对“虚拟世界”的概念定义,我们指出“虚拟世界”的至美境界就是现实世界完整而生动的“映像”,建立虚拟世界的目的不仅在于对虚拟事件的时间轴推演与回顾,更重要的建立一个可以修复的世界去供我们破坏,从而为现实世界中的各种人类活动、科学实验、环境评估提供一个生动的实验平台。作为主要的背景知识,本文对GPU功能结构及GPU编程技术方面作了详细的论述。针对GPU硬件体系快速的性能提升与功能拓展,我们简述了各个时期GPU的功能特性与GPU性能的评价标准,将GPU的可编程性能确立为新的GPU性能评价标准,足显GPU作为新一代图形系统开发的必然发展趋势;同时我们还结合三维图形系统开发过程介绍了新旧两种三维图形系统渲染方式,指出了GPU图形系统开发的强大功能支持、快速渲染与灵活实现等众多优势;GPU编程是一种与传统算法编程思想完全不同的机制,它主要面向顶点数据与像素数据进行编程,其实现建立在大型点阵的快速并行运算技术上,作为本文实现的主要技术基础,我们还简要地介绍了GPU可编程语言的文法与编程思想。在针对虚拟地理环境的研究中,我们的研究内容主要集中在大区域地形可视化、光照渲染与水体可视化三个方面。在大区域地形可视化技术方面,分为三个步骤来论述:模型表达、纹理生成与渲染机制。主要研究围绕着如何解决在大区域表达时的实时效率与可视化品质相矛盾的问题,解决办法不外乎在各个环节上都注重效率的提升。为此,在模型表达中,我们的解决思路有两条:一、对传统的地形表达算法进行大区域地形可视化调度策略上的优化;二、运用GPU建立没有模型数据结构的快速渲染机制。针对第一条技术路线,我们论述了基于二叉树的双缓冲区队列实现的实时优化自适应网格模型(Roam算法)的实现,同时引入了一种先进的能够符合大区域地形分块调度的地形场景分页管理器(PLSM)机制。由于PLSM采用事件监听驱动方式、具备独立GPU渲染方式、模拟操作系统高效调度策略等众多优点,非常适合大区域地形的调度表达,因而我们将Roam算法与PLSM结合起来实现了大区域的高效渲染:Roam算法作为三角网建模算法直接负责了地形的LoD建模,而PLSM则负责包括地形块调度、数据接口以及独立的渲染机制等诸方面的工作。在第二条技术路线实现方面,文章论述了GPU实现的地形表达经典算法,美国斯坦福大学与微软研究院共同研究的科学成就----基于几何裁剪的嵌套网格模型算法的实现机制。这个算法的优点就是不需要建立复杂的地表模型,几乎所有的工作均可以在GPU中进行高效处理,不需要过多地占据CPU的计算资源,是一种非常巧妙的算法实现;结合本人先前的关于同层多分辨率地形表达模型的研究,本文对这一模型提出了自己的改进思维,建立了综合网格嵌套的多分辨率表达模型。主要的改进是基于视向与视距来确立多分辨率的层次分布从而实现精度控制的LoD地形模型表达。在纹理生成方面,针对大区域地形表达的纹理获取困难、匹配困难、多分辨率表达困难等原因,文章论述了纹理投射(Texture Splatting)的动态纹理生成方法,这种方法的好处不仅在于可以得到时态纹理,而且可以根据纹理成分合成纹理,从而在没有真实纹理的情况下可以根据各种纹理成分的分布图来生成具有相当仿真程度的匹配纹理;本文同时也基于纹理投射的原理提出了改进的,能够显著提高渲染效率,并且不影响动态纹理表达灵活性渲染的补偿纹理合成方法。这一方法的思想是将纹理投射过程分为两个过程,在前期实现静态的纹理成分合成预处理,在后期针对时态性的纹理(如雪、水面)等实现实时的纹理成分补偿性的叠加,并且将光照计算的部分工作也放在了纹理动态合成的环节中,从而进一步提高了大地形表达的实时效率。在渲染机制方面,传统的管道方式在要求实时性非常高的大区域渲染方面基本上已经退出了历史的舞台,在大型的三维可视化系统中,均采用了GPU独立渲染的表达机制。运用GPU进行渲染,可以完全从地形建模中独立出来,而且还可以对同一模型定制多套渲染方案,渲染方案的描述也非常简单,通过渲染脚本描述,并由三维引擎直接送到GPU进行渲染。文章对这方面的知识进行了详细论述。在虚拟地理环境的自然光照效果方面,我们主要从物理光学方面由简入繁、深入地分析了光照合成的基本原理,并用物理的光学分析实现了对光盘衍射干涉光波的效果模拟,深入到微观的物理粒子世界去探析自然规律与自然现象的统一。文章从管道渲染方式的基本的光照模型、到聚光灯效果、光衰减效果的扩展模型,再到环境映像贴图的技术实现,通过GPU的编程技术实现,充分地体现了GPU编程的灵活性优势与强大功能拓展,为实现复杂的光效渲染提供了支持。作为对这一观点的验证,我们通过物理光波的粒子跟踪,模拟实现了光盘光波干涉现象的模拟。在水面的表达模型上,我们综述了水面模型的三种表达方式:纹理波、几何波与物理波的技术实现。在纹理波的表达中,针对静态网格数据的放射状采样LoD表达提出了更为科学的、根据视线与视点位置确立的椭圆采样表达模型,这一模型同样可以应用在Grid-DEM表示的地形模型中;在水面表达的纹理波叠加方法中,我们提出了根据信号脉冲方式驱动的纹理波叠加与置换方法,并且给出了运用GPU编程实现了的纹理叠加算法。在水面的几何波的表达中,我们运用地图代数距离变换方法以纹理图的方式给出了不规则形体辐射波的方向定义,并且能够在障碍水域中同时表达多个波源的辐射波作用;同时我们还讨论了常用的基于快速离散傅立叶变换(FFT)的几何波采样算法实现。在水面的物理波表达中,结合流体动力学表达的Gerstner波,我们通过融合更多的物理参数,实现了由数学模型到物理模型的转变,演绎出外力作用下的物理波表达模型,并且在GPU中巧妙地实现了这种模型的表达。作为虚拟自然环境的另外一个重要研究主题,虚拟物理环境是一种最为抽象的技术实现。它主要是提供一种物理空间表达机制,在这种机制下,实体的行为被动地受到空间机制的约束从而表现出相应的物理运动。在这一个方面,本文主要论述了当前的物理加速硬件实现技术,并且重点以Newton物理引擎为例揭示了物理引擎机制的技术实现,分析了物理引擎的功能结构与技术架构,同时我们还以大地碰撞模拟实验为例讨论了物理引擎的应用。最后,我们还以GPU编程为实现手段,通过对皮球的弹跳模拟实现反映了GPU对物理运算及物理现象模拟的支持能力。在虚拟气象环境表达中,本文总结了建立虚拟气象表达模型的几种技术路线,并且对技术非常先进的直接体视化技术(Splatting)进行了详细的介绍。而针对常用的粒子系统表达方式,我们提出了面向粒子的GPU通用表达统一模型,融合粒子绘制、运动更新、光照渲染表达于一体,同时可处理静态粒子集合随机采样粒子与依据气象模型动态实时生成的粒子,为虚拟气象现象模拟提供了统一的GPU编程实现模式。在本文的最后,我们针对虚拟自然环境在技术与实现上的综合,从主题整合、效果融合以及技术体系的协作三个方面进行了探讨,其内容包括:VGE、VPE与VME的关系布局;图形渲染、光照效果、物理计算的效果融合;CPU、GPU与PPU的技术协作等内容。综观全文,本文紧紧围绕着虚拟自然环境的各个分支主题的模型表达以及GPU编程模拟技术两个方面展开论述。作为本文的研究主旨,我们的研究主要在于探讨建立虚拟自然环境的技术路线、体系架构以及模型理论综合等,为建立虚拟自然环境系统进行一些铺垫性工作。

【Abstract】 GPU’s rapid development beyond the speed of the semiconductor technology Moore’s Law three times provides an unprecedented new tool and technology for the development to achieve large-scale real-time 3D simulation system, which will greatly promote cognitive enhancement of people’s awareness of the 3D simulation system. Virtual Natural Environment surpasses the expression of entities in VR, it mainly includes the Virtual Geographic Environments(VGE),Virtual Physical Environment(VPE) and Virtual Meteorological Environment (VME) and many other abstract but objective themes, which depicts a greater sense of immersion and more comprehensive expression of the space environment.Based on integration of current domestic and foreign advanced technology, combining advanced GPU programming, in this paper we strive to build a complete technology system of Virtual Natural Environment implementation. The content of this paper focus on model building and GPU’s implement of VGE, VPE and VME’s field. The chapters are organized independently and the topic is clear, but all are related with research content and realization of VNE.In chapter 1, we discuss the development of spatial cognition and regard technology development and means updating as the important factors influencing spatial information expression. Based on this and combining the great updating in current performance of advanced GPU graphics developing, we launch the cognition system structure of“Virtual World”. Then we point out that virtual world can be divided in“VNE”(Virtual Natural Environment) and“Virtual Society”, and the division standard depends on“whether there are people participating or not”. Then we analysis the constituent system of VNE, confirm the three fields of VGE, VPE and VME as the content of VNE and discuss the content and technology realization of each theme. With regards to Virtual Reality and Virtual World, we discuss the relation between virtual reality technology and virtual world (including its researching branches) and make it clear that“Virtual World”is the goal while virtual reality technology is only a researching means ,a realization way. Finally we point out that virtual world’s perfect state is the full and vivid image of real world, the purpose of building virtual world are to create a restorable world for us to destroy more than to deduce and review the time frame of things. So that it can provide human activities, scientific experiment and environmental assessment with a lively experiment platform.As the major background, the functional structure of the GPU and GPU programming aspects are discussed in detail in this paper. With regard to GPU’s hardware performances upgrade and function expansion, we outline GPU’s functional properties and evaluation standard of the performances in different period and establish GPU programmability as the new evaluation standard of GPU performances. It highlights that GPU will be the inevitable developing trend in the new generation of graphic system development. We also combine 3D graphic system development and introduce both old and new 3-D graphics system rendering. Then we point out several great advantages of GPU graphic development in strong function support, fast rendering and flexibility. GPU programming is of the mechanism that differs from the traditional algorithm idea and it mainly gears to the pixel and vertex data to program .Its realization relies on rapid parallel computing technology of large array. We also brief GPU programmable language grammar with programming ideas as the main realization basis.In the research on virtual geographic environment, our main content focus on regional terrain visualization, illumination rendering and water visualization.On regional terrain visualization technology, there are three themes to address: model expression, texture creation and rendering mechanism. The main research focuses on how to work out the contradiction between real-time efficiency and visual quality in regional expression. We have two resolution ways: the first one is to optimize the regional visualization strategy in traditional terrain express algorithm. The second is using GPU to build rapid rendering mechanism without model data structure.Concerning the first technology, we discuss the real-time optimization automation model (Roam) whose realization bases on the bi-buffer queue of the binary tree and introduce the advanced PLSM that can comply with the regional block scheduling topography. As PLSM has many advantages, such as its incident wiretapping drive, independent GPU rendering ways, efficient dispatching strategy of simulating operation system, it fits regional terrain dispatch and express. So we combine Roam algorithm and PLSM to implement efficient regional rendering. Roam algorithm, as triangulation modeling algorithm, is directly in charge of LOD modeling while PLSM is in charge of terrain attempering, data interface and independent rendering mechanism etc.Concerning the second technology, this paper discusses the classic terrain express algorithm realized by GPU. That is the common scientific achievements of Stanford University in the United States and Microsoft Research Institute----Nested-grid model algorithm implementation mechanism based on the geometric cut. This algorithm’s advantage is that it doesn’t need to build complex terrain model and almost all the work can be done efficiently in GPU at the cost of occupying more GPU computing resources, thus it is a clever algorithm. Combining my previous research on terrain express model based on the same level with multi-resolution, I launch some improving ideas of this model and establish the multi-resolution model terrain expression of integrated nested grid in this paper. The main improvement is to establish level distribution of multi-resolution based on horizon and distance and thus to implement the LOD terrain model express of precision control.On texture creation, as there are difficulties in texture acquiring, matching and multi-resolution express of regional terrain express. In this paper, a dynamic texture creation method of Texture Splatting is discussed. The advantage of this method is that we can not only get dynamic texture but also synthesize texture with its constituent. Then we can get matching texture with considerably good simulation without real texture by texture distribution map of its various components. This paper also launches an improved compensation texture synthesis way which is based on Texture Splatting , can increase rendering efficiency and would not affect the rendering flexibility of dynamic texture expression. The idea of this approach is to divide the procedure into two periods. In the former period, static texture component synthesis pre-processing is finished. In the latter one, real-time texture component compensation overlaying of tense texture(like snow, water surface) is completed and part of illumination computing is done in this period. In this way, real-time efficiency of regional express is greatly enhanced.On rendering mechanism, the traditional pipeline has to withdraw from the historical stage at the real-time requirement of large area rendering. In large 3-D visualization system, GPU independent rendering express mechanism is adopted. Using GPU, rendering can be independent from terrain modeling and a set of customization programs can be created for one model. It is very simple to describe the rendering program. With rendering script, the program can be sent to GPU by 3-D engine. Knowledge on this aspect is introduced in detail in the paper.On natural lighting effects of VGE, we mainly analysis the basic principles of light synthesis from simple to complex and in depth from the physical optics aspect and achieve the effect imitation of CD diffraction interference wave using physical optics analysis. This goes deep into the micro physical-particle world to explore the unification of natural discipline and natural phenomena. In this paper, first we introduce the expansion models of illumination model, spotlight effect, optical attenuation effect which comply with the pipeline way, then the technology accomplishment of environment image mapping. The advantage of GPU programming flexibility and strong function expansion can be full embodied through GPU programming technology. And this supplies help for complex light rendering realization. As certification of this point, we complete wave interference simulation by physics beam particle tracking.On water surface expression model, we review three expressions: texture wave, Geometry and physics-wave technology. On texture wave expression, according radical sampling LOD expression of static grid data, we put forward elliptical sampling expression model which is more scientific and is confirmed by view sight and location. This model can also be applied to terrain model of Grid-DEM expression. In texture wave superposition method of water surface express, we put forward texture wave superposition and replacement method through signal pulse-driven and texture superposition algorithm using GPU programming express. In water surface geometric express, we define directions of irregular shape radiation wave in the way of texture wave using distance transform of map algebra and express wave radiation of much wave resources in water with obstacles. At the same time, we discuss the common algorithm of geometric wave sampling based on fast Fourier transform (FFT). In water surface physical express, combining Gerstner wave in fluid dynamics, we realize the transform from mathematical model to physical model and deduce physical wave express model in outside force and finally accomplish the model expression ingeniously in GPU.As another important theme of VNE, virtual physical environment is an abstract technology realization. It primarily provides a mechanism of physical space expression. With this mechanism, entities’actions passively display corresponding physical movement under the space constraint. In this aspect, this paper mainly discusses current physics acceleration hardware technology line and demonstrates physical engine mechanism realization by the example of Newton physical engine. Then we analysis function structure and technology framework of physical engine and exemplifies the experiment of collision with the earth to discuss the physical engine use. Finally, with GPU programming as tools, we use the bouncing ball simulation to reflect the support capacity of GPU for physics computation and physical phenomena simulation.In virtual meteorological environment expression, this paper summarizes several technology ways of creating virtual meteorology expression and introduces the advanced directly Visualization Technology (Splatting) in detail. As for the common particle system expression, we launch particle-oriented GPU unified model with common express which integrates particles drawing, sports updates and light rendering expression. Also this new model can deal with real-time and dynamic generating particles with random sampling static particles and meteorological model. This provides a united realization mode of GPU programming for virtual meteorological phenomena simulation.In the final part of this paper, focusing on comprehension of technology and realization of VNE, we discuss theme integration, effects integration and technical cooperation. The content includes relation layout of VGE、VPE and VME; Graphic Rendering, light effects, effect integration of physical calculation,CPU, GPU and PPU technical collaboration and so on.From the whole view, this paper discusses the model expression of all branches of VNE tightly and simulation technology of GPU programming. As the main purport of this paper, our primary task is to explore the technology line of establishing virtual natural environment, the architecture and integrated support of the model.

  • 【网络出版投稿人】 武汉大学
  • 【网络出版年期】2008年 07期
节点文献中: 

本文链接的文献网络图示:

本文的引文网络