>> 杜志强，2007，感知驱动的复杂建筑物模型简化方法
摘 要
虚拟地理环境描述的是现实世界中无限复杂的地理特征和地理现象，如果不根据具体的应用需求对三维地理特征和现象进行有效地抽象或者简化，那么数据规模将会无穷大。此外，在实际生活中，人眼的空间与时间分辨率总是有限的。根据这一基本原理，应该忽略掉那些人眼所不能察觉的空间物体细节，并在此基础上，完全可以通过有限的若干离散的细节层次就能获得有关建筑物特征及其地理环境比较完整的知识。自动获取满足视觉连续性要求的离散的多细节层次模型成为降低交互式建模强度的关键因素和虚拟地理环境实时可视化的重要基础之一。本文在详细分析了传统的简化方法基础上，针对复杂建筑物的特点，为了降低问题的复杂性，直接从人眼感知特性出发，充分利用计算机屏幕图像信息解析感知特征，并利用感知知识准确定位待简化的部件、图元，从而为模型简化提供引导知识。本文的研究内容包括：
（1）针对虚拟地理环境的核心数据要素——复杂建筑物三维模型的特点，深入剖析一般网格模型自动简化技术，针对建筑物模型的自动简化技术以及现有的简化工具，综合评价了各种自动简化技术和简化工具的研究进展和存在的不足之处，归纳提炼出三维复杂建筑物模型自动简化技术的发展趋势。
（2）针对难以准确定位待简化的部件、图元以及细节层次的问题，提出了感知驱动的复杂建筑物三维模型简化方法。感知驱动是复杂模型自动简化特别是连续简化的最原始动力，也是最难以量化控制的。基于人眼从屏幕图像获取有关三维模型感知信息的基本事实，本方法从模型的渲染图像入手，分析了渲染图像生成的各种要素，提出了渲染图像生成与分析方法，并根据人眼感知图像的原理通过模拟视觉图像与原始渲染图像之间的差异利用小波分析探察人眼不可感知的细节。开发了图像像素亮度反映出的感知度差异聚类与映射算法，将图像像素的感知度信息传递给三维模型，从而利用图像感知知识直接驱动模型简化操作。
（3）提出了三维复杂建筑物模型自动简化算法。其中主要包括三个方面的内容
研究了适合于部件组成的复杂建筑物模型简化操作的半边数据结构，并在此基础上提出了半边数据结构中的语义扩展，有利于在简化模型过程中直接利用模型特有的语义信息作为空间剖分的依据，对空间离散的部件模型进行数据重组，有效地克服了无组织模型数据的缺点。此外，在简化后可以将模型的几何、属性以及语义信息一并导出，克服了简化模型中难以保持语义信息的缺陷。
提出了基于语义的部件模型简化方法。复杂建筑物模型是由众多彼此之间相对独立的部件构成，而单个部件模型包含了大量用于相互接驳的凹陷特征。面对这些特征，本文提出了基于离散符号语义描述特征的方法，确定顶点层级与层级跃进的规则，通过顶点层级跃进和半边折叠简化操作有效地简化了部件上全部的凹陷特征。
开发了感知驱动的模型简化算子以及保持外观属性的算子。在简化操作方面，重点研究了新顶点的生成方法、点对合并的代价，提出了以顶点亮度变化程度作为点对合并代价的感知权重，通过感知权重调整点对排序，进而使得简化模型在几何误差和感知误差之间得到了平衡，克服了以前纯粹以几何代价确定简化操作的不足。在外观属性保持方面，深入分析了几何操作引起外观属性变化的各种情况，有针对性地提出了关于纹理、材质等方面的保持外观属性的算子。
在上述研究成果的基础上，采用完全自主知识产权的VGEGIS软件技术，以一个中国古典建筑为案例，在微机平台上构建了建筑物自动简化原型系统，验证了本文研究成果是切实可行的，在能够大幅度降低复杂建筑物模型的复杂程度的同时，使得简化模型很好地保持了原始的外观特征。通过提供多种参数选择，该原型系统不仅能够按照用户指定的感知简化策略执行各种不同等级的模型简化，创建出满足用户需要的简化模型，而且可以按照用户指定的简化阈值进行简化，为虚拟地理环境建设中的复杂建筑物LOD建模提供了一种新工具。
关键词：
虚拟地理环境，LOD建模，复杂建筑物三维模型，视觉感知，自动简化，半边数据结构，语义
Abstract
Virtual Geographic Environments (VGEs) are designed to describe unlimited complex geographic features and phenomena. If we do not effectively abstract and generalize the 3D geographic features and phenomenon according to application requirements, the data volume will be infinite. Moreover, the spatial and temporal resolution of the eyes is finite in real life. Based on this fact, the details not be perceived by human eyes should be ignored in modeling process. It is possible to obtain the building features and entire acknowledge of geographic environment through the finite and discrete LOD models (LODs). Creating the LODs that can meet the requirement of visual continuity becomes a key element for dramatically decreasing the intensity of manual modeling and it is one of the important bases for real-time visualization. Therefore, how to effectively reduce the redundancy of 3D data model becomes the basis of 3D modeling, real-time visualization, and high-efficient spatial analysis. It is a problem that needs to be resolved in 3D Geographic Information System (GIS).
Complex building models are widely used in various fields, such as VGEs, cyber city, and culture heritage documentation, with well development of the virtual reality and 3DGIS technologies. A complex building model means it has highly topological and geometric complexity. In order to obtain such complicated structure and precise texture, the model usually assembled by more than ten thousand pieces of components. But such high complexity and huge data volume hamper the real-time and interactive application requirements of complex building models in virtual environments.
Aiming at the characteristics, the traditional simplification methods have two drawbacks. One is that the most traditional simplification methods in computer graphics are restricted to simplify a single 3D object with continuous surface mesh but not a set of meshes, but a complex building model is just a set of mesh because each component is a mesh. The other one is due to the simplification methods do not take human perception information to drive the simplification operations, so that the simplified results can not accord with the rule of human perception, and the LOD models derived from the simplification methods are difficult to ensure the continuity on visual effect. The main contents of this research include:
(1) After reviewing the various automatic simplification methodologies, technologies, and implements, the existing simplification methodologies are analyzed. Moreover, the states of art and the development tendency of the technologies and tools are indicated.
(2) Aimed at the characteristics of complex building models, perception-driven simplification methodology is proposed, which means interpreting the perceptual features of the image displayed on the screen, and locating the component and geometric primitives that will be decimated, accordingly, providing pilot acknowledge for simplification operations. This methodology includes five main steps, and simulation of the perceptual image on retina, clustering according to the perceptual difference of pixels, and projection of the perception information are the key points in the five steps.
(3) Perception-driven simplification algorithm, which includes three aspects:
A new half-edge data structure is proposed, which can meet the requirement for the simplification operation of the complex building. In this research, the data structure is expanded on semantic information, so that the semantic information are used as foundations of spatial partition and this can effectively overcome the unstructured states among building components. Furthermore, the semantic information kept in the simplified models can serve the spatial attribute query in VGEs.
Semantic-based simplification for building components can remove the concave features on the surface of component models. After discrete signal method is used to describe the surface features of component models, levels of vertex and the rule of vertex transition lead to half-edge collapse operations until the all the concave features are eliminated.
Designing operators for perception-driven simplification and for preserving the appearance attributes are main contents in this algorithm. On the simplification operator, creating new vertex and calculating the cost of vertex pair contract are key elements. For overcoming the drawback of the pure geometrical cost, calculating the intension change between new vertex and original vertex can be weighted with the geometrical cost. This method helps to balance the geometrical error and perceptual error. On the appearance attributes, the operator for preserving appearance attributes is developed, aimed at various changes of appearance attributes caused by geometrical simplification.
Based on the theoretical results and VGEGIS software made by LIESMARS, taking a typical Chinese ancient building as a case, a prototype of automatic simplification for complex building models is built on personal desktop computer, which can prove that these research works are feasible. This prototype can not only dramatically reduce the number of triangles, but also keep the appearance of models well. Moreover, the prototype provides multi-levels simplification to create LOD models according to users’ conditions, such as hardware, system configuration, for scarifying their requirements. It facilitates creating LOD models of complex building in Virtual Geographic Environments.
Key words:Virtual Geographic Environments，LOD Modeling，Complex 3D Building model，Visual Perception，Automatic Simplification，half-edge data structure，semantics
