【作者】 淮永建；

【导师】 郝重阳；

【作者基本信息】 西北工业大学 ， 信号与信息处理， 2002， 博士

【摘要】 虚拟现实技术在飞行仿真、交互式GIS、虚拟购物、数字地球和游戏娱乐等领域都有广泛的应用。这些应用不仅要生成高度真实的复杂虚拟环境，而且用户要和虚拟环境进行互动，增强交互性和沉浸感。使人可以随心所欲的在环境中漫游，跟其在真实的环境中体验一样。随着虚拟现实技术应用的深入，人们对虚拟场景的复杂度和真实感要求越来越高，远远超过了计算机图形硬件实时处理能力。因此迫切需要解决场景的复杂性和交互实时性这一矛盾。虚拟场景的实时生成是虚拟现实技术中极富挑战性的课题和研究领域。 地形、云彩以及自然景物(如树木)和其它复杂几何模型是虚拟场景中的重要组成部分。论文对虚拟场景中的自然景物、场景中复杂几何模型的实时绘制技术、运动物体的实时建模技术等进行了深入研究，取得了一些研究成果，并将这些成果成功应用于虚拟场景的实时绘制中。在提高实时性的前提下，尽可能满足视觉逼真度的要求。 论文取得的研究成果如下： 1．提出了一种基于分形随机中点位移方法的视相关的多分辨率层次细节LOD实时地形简化算法。首次将层次细节LOD技术引入到分形地形绘制中，并将该方法成功应用于三维的云彩绘制，利用随机函数特性模拟云彩的动画。依据视点的变化选择适当分辨率的地形细节进行绘制，当视点靠近地形分维曲面时，可以局部地增加细分层次以获取更多表面细节；而当观察者远离分维曲面时，则可减少分割层次来节省计算量，提高实时绘制性能。未引入层次细节简化前的分形地景的绘制速率为35FPS，引入后的地形绘制速率为65FPS，绘制实时性明显提高。 2．提出一种基于分形迭代L～System的视相关三维树实时绘制算法。自然界中许多植物的整体和局部具有很强的自相似性，而且这些植物特征复杂，很难通过传统的建模方法进行绘制。论文分析了迭代L～System在模拟三维植物中的优势，并利用它生成离散的多层次细节LOD三维树，实时绘制时，依据视点距离树的位置，选择适当分辨率层级的LOD树进行绘制。在树叶模拟时，提出一种植物的形态构造法进行模拟。最后将三维分形树集成到利用随机中点位移法生成的分形地景中，地景绘制的帧速率为46FPS，可实时进行交互浏览整个场景。 3．提出一种视相关连续的多层次细节LOD复杂多边形几何模型的简化算法。算法基于边折叠的思想可生成具有任意细节度的LOD模型。在折叠边的代价估计中，均衡考虑了多边形边的长度和多边形表面曲率影响；折叠边的顶 西北工业大学博士学位论文点位置从折叠边的顶点中选取，既有助于保留了初始模型的形状，又减少运算量，满足快速绘制的需求。在依据视点选择适当LOD模型时，采用了顶点渐变（w巾x＊。叩hn幻技术光滑多边形模型，减少不同层次＊OD模型转化时引起的图像‘跳跃’。利用该算法对复杂儿何模型进行简化，在满足视觉逼真度的前提下绘制帧速率可高达90FPS左右，比简化前提高了30FPS。 4．提出了一种基于自适应四叉树结构，视点相关的动态多分辨率层次细节LOD实时地形绘制算法。主要讨论了地形简化中多分辨率地形网格生成、节点评价函数、多分辨率LOD转化时的图像跳跃以及多分辨率地形纹理映射等关键技术问题，有效地解诀了多分辨率地形网格拼接中的裂缝问题，利用顶点渐变技术光滑不同层次细节转化时的图像跳跃现象。首次讨论了高细节度实时地形绘制中的纹理简化技术。对大的、高细节地形纹理图像，通过纹理图像金字塔和纹理四叉树来构造不同分辨率的纹理图像。实际绘制时，从纹理树中选择与地形几何层次细节相一致的纹理分辨率图像进行映射。从而有效的解决了大地形纹理图像的映射时间和存储问题。最后，将该算法应用于DEM地形模型简化中，简化前的地形绘制帧速率为4FPS，简化后的地形绘制帧速率为70FPS左右，在绘制图象质量没有明显退化的前提下，能满足视觉和实时划览的需要。 5．提出了一种面向虚拟环境实时应用，基于多体动力学理论通用的实时车辆建模方法，克服了以往动力学车辆建模中“一车一模”的弊病。该方法通过车辆的动力学和运动学分析，将车辆看成是由多个自由度的刚体组成，依据实际的受力状态，建立车辆的通用动力学方程。既能逼真的模拟车辆在虚拟场景中的各种运动行为和姿态，又能获得很高的帧速率，整个仿真场景的帧速率为40FPS左右，满足车辆驾驶中实时交互性的要求。更多还原

【Abstract】 Virtual Reality has been widely used in the fields such as flying simulation,interactive GIS,virtual touring,digital earth,information visualization,architecture walkthrough,etc. These application need not only realism complex Virtual Environment,but require the user can interact with the virtual scene at interaction frame rate. As long as meeting the real-time rendering,the user can feel the virtual environment is the same as the real world and immerse into the environment. However,Virtual reality communities and other visualizations have always faced the problem that their "desirable" visualization dataset sizes are one or more orders of magnitude larger than what the hardware can display at interactive rates. Thus we need resolve the problem by attempts to bridge the gap between the desired and the actual hardware performance,through algorithmic and software techniques. The most critical problem in VR is real-time rendering the complex Virtual Environment.The general virtual environment is composed of terrain,clouds,tree,forest and other complex geometry models. In this dissertation,we mainly research the real-time rendering of complex virtual environment which includes the real-time terrain rendering,real-time modeling,3D complex model simplification and acceleration techniques. We have gained some research results and realize these innovation to actual virtual scene rendering which can meet the requirement of interaction without clearly degenerating image quality.The main research results include:1. In this dissertation,we present an algorithm of view-dependent continuous level of detail terrain simplification and 3D clouds rendering based-on fractal. The terrain mesh generated by using previous random mid-point displacement is usually uniform. When the viewpoint is far away from the terrain block,you need not rendering the terrain with higher detail. We introduce the method of LOD simplification into fractal terrain rendering to generate multiresolution terrain and reduce the triangles rendered and improve the performance. And,we have successfully used this method to render three dimensions clouds. You can dynamically simulate the motion of clouds by using the random function and adjust the random displacement of terrain height value and fractal dimension to change the terrain silhouette and surface roughness. In order to resolve the clear terrain crease,we use vertex interpolation to smooth the height value and enhance the coherence between of vertexes.2. In this dissertation,we present an algorithm of view-dependent LOD three dimension fractal trees. The tree modeling is based-on fractal iterated. The algorithm can simulate different kinds of trees and provides a method of leaf simulation that is named plant shape construction. The algorithm has many advantages such as simple,easy to carry out,a little calculation and a good real-time characteristic. By controlling the iterated levels,you can automatically generate LOD trees and select the appropriate LOD model to render according to the viewpoint changing.3. In this dissertation,we present a new algorithm of polygon simplification togenerate different LOD based on the edge collapse. When calculating the cost ofedge collapse, we consider the influence of edge length and surface curvature. And,the position of the new vertex is selected from the veYtexes of the edge collapsed,this can not only preserve the initial model shaPe, but also reduce the calculation ofCPU and meet real-time rendering of Virtual Environment. Vertex morphing is usedto smooth LOD model to reduce the popping between different LOD transition. Wehave enhanced the polygon data structure to implement the algorithm and proved theefficiency of the algorithm by a typical example.4. In this disseYtation, we present a new algorithm of View-Dependentmultiresolution terrain simplification based on adaPtive quadtrees. Especially, wediscuss these main problems in multiresolotion terrain rendering such as levels ofdetail mesh generation, ved更多还原