Collaborative Research in Computer Graphics: Real-Time Visualization and Rendering of Complex Scenes

本文主要研究复杂光照、真实感材质、光散射的精确模拟以及交互式参数处理等实时绘制问题。本文探讨了外观表示的基本问题，并提出了一种新的实时绘制算法。本文主要研究了以下三个方面：（i）对外观表示和光传输的复杂性进行数学和经验分析;（ii）计算技术，如从细到粗和从粗到细的算法、函数字典和分层主成分分析;以及（iii）支持照明、材料和视点的交互式操作的算法和系统。计算机图形学和科学可视化越来越多地用于诸如模拟和训练、汽车设计、建筑渲染、国防、电子商务和娱乐。在今天的技术下，使用复杂但缓慢的渲染算法来创建惊人的逼真图像是可能的。这些算法不能满足日益增长的对有效方法的需求，以操纵和与3D环境交互。现代PC可以以交互速率渲染具有数百万个多边形的场景，但这种性能是以质量和真实感的相当大的损失为代价的。这个项目试图克服这些限制：以交互的速度产生高度逼真的图像。在过去的三十年里，计算机图形学的研究已经创造出了提高计算机生成图像的真实感和生成速度的技术。 然而，没有广泛使用的方法来创建逼真的图像，并且在实践中创建逼真的图像仍然很困难。 该项目旨在开发一种实用的方法来创建逼真的图像，以解决图形和感知社区认为突出的视觉效果。 该方法的目的是强大的，相对简单，并有可能在未来十年内是互动的。快速生成逼真的图像的能力是有用的，在许多应用。 例如，飞行和驾驶模拟器依赖于逼真的图像，但这些系统的交互性限制了可行的真实性。 拟议的工作应扩大这些限制。 所提出的方法也将有助于生成规划和教育的图像;图像可以帮助可视化重建工作，生境变化和城市规划的结果。 最后，该方法应该是有用的，用于生成输入到自动车辆，其中许多不同的场景和大气条件可以快速生成，成本低的合成图像。

ABSTRACTReal-time rendering with complex illumination, realistic materials, and anaccurate simulation of light scattering, with the ability to manipulateparameters interactively, is the challenge addressed in this project.Fundamental questions about the representation of appearance are explored,resulting in the development of a new class of real-time algorithms.Specifically, the focus is on three main directions: (i) Mathematical andEmpirical Analysis of the complexity of appearance representations andlight transport; (ii) Computational Techniques such as fine-to coarse andcoarse-to-fine algorithms, function dictionaries, and hierarchicalprincipal component analysis; and (iii) Algorithms and Systems supportinginteractive manipulation of illumination, materials, and viewpoint.Computer graphics and scientific visualization is increasingly used inareas such as simulation and training, automobile design, architecturalrenderings, defense, electronic commerce, and entertainment. With today'stechnology, it is possible to create strikingly realistic images usingsophisticated, but also slow rendering algorithms. These algorithmscannot address the growing need for effective methods to manipulate andinteract with 3D environments. Modern PCs can render scenes withmillions of polygons at interactive rates, but this performance comesat the cost of a considerable loss in quality and realism. This projectseeks to overcome these limitations: producing highly realistic imagesat interactive rates.Over the last thirty years computer graphics research has created technology that improves both the realism and generation speed of computer-generated images. However, there is no widely used method for creating realistic images, and it remains difficult in practice to create realistic images. This project aims to develop a practical method for realistic image creation that addresses the visual effects believed to be salient by the graphics and perception communities. The method is intended to be robust, relatively simple, and to have the potential to be interactive in the next decade.The ability to rapidly generate realistic images is useful in many applications. For example, flight and driving simulators rely on realistic imagery, but the interactive nature of these systems limits the realism that is feasible. The proposed work should extend these limits. The proposed method would also be useful for generating images for planning and education; images could be created to help visualize the results of reconstruction efforts, habitat change, and urban planning. Finally the method should be useful for generating synthetic images for input to automated vehicles, where many different scenes and atmospheric conditions could be generated rapidly and at low expense.