三维建模毕业答辩论文英文文献翻译.doc

三维建模毕业答辩论文英文文献翻译

毕业设计(论文)
文 献 翻 译

题 目 矿山测量三维模型制作
学 院 测绘科学与技术学院
专 业 测绘工程
班 级 0802班
学 生 史姝
学 号 0810020205
指导教师 汤伏全

二〇一二 年 六 月 二 日

Constructing Rules and Scheduling Technology for 3D Building Models

Zhengwei SUI, Lun WU, Jingnong WENG, Xing LIN, Xiaolu JI

Abstract

3D models have become important form of geographic data beyond conventional 2D geospatial data. Buildings are important marks for human to identify their environments, because they are close with human life, particularly in the urban areas. Geographic information can be expressed in a more intuitive and effective manner with architectural models being modeled and visualized in a virtual 3D environment. Architectural model data features with huge data volume, high complexity, non-uniform rules and so on. Hence, the cost of constructing large-scale scenes is high. Meanwhile, computers are lack of processing capacity upon a large number of model data. Therefore, resolving the conflicts between limited processing capacity of computer and massive data of model is valuable. By investigating the characteristics of buildings and the regular changes of viewpoint in virtual 3D environment, this article introduces several constructing rules and scheduling techniques for 3D constructing of buildings, aiming at the reduction of data volume and complexity of model and thus improving computers’ efficiency at scheduling large amount of architectural models. In order to evaluate the efficiency of proposed constructing rules and scheduling technology listed in the above text, the authors carry out a case study by 3D constructing the campus of Peking University using the proposed method and the traditional method. The two results are then examined and compared from aspects of model data volume, model factuality, speed of model loading, average responding time during visualization, compatibility and reusability in 3D geo-visualization platforms: China Star, one China’s own platform for 3D global GIS manufactured by the authors of this paper. The result of comparison reveals that models built by the proposed methods are much better than those built using traditional methods. For the constructing of building objects in large-scale scenes, the proposed methods can not only reduce the complexity and amount of model data remarkably, but can also improving computers’ efficiency.

Keywords: Constructing rules, Model scheduling, 3D buildings

I. INTRODUCTION
In recent years, with the development of 3D GIS (Geographical Information System) software like Google Earth, Skyline, NASA World Wind, large-scale 3D building models with regional characteristics have become important form of geographic data beyond conventional 2D geospatial data, like multi-resolution remote sensing images and vector data [1].Compared to traditional 2D representation, geographic information can be expressed in a more intuitive and effective manner with architectural models being modeled and visualized in a virtual 3D environment. 3D representation and visualization provides better visual effect and vivid urban geographic information, and thus plays an important role in people's perceptions of their environment. Meanwhile, the 3D building data is also of great significance for the construction of digital cities.
But how to efficiently visualize thousands of 3D building models in a virtual 3D environment is not a trivial question. The most difficult part of the question is the conflicts between limited processing capacity of computer and massive volume of model data, particularly in the procedure of model rendering. Taking the 3D modeling of a city for the example using traditional 3D modeling method, suppose there are 100 000 buildings to model in the urban area and the average size of model data for each building is roughly 10 M. So the total data volume of building models in the city could reach a TB level. However, the capacity of ordinary computer memory is only in the GB scale. Based on this concern, the authors proposed the scheduling technology for large-scale 3D buildings models in aspects of model loading and rendering. Due to the lack of building constructing rules and standard, models of buildings vary in aspects of constructing methods, textures collection and model data volume, especially in aspects of model reusability and factuality. Such a large amount of data without uniform constructing rules becomes a huge challenge for data storage, processing and visualization in computers. It also brings the problem of incompatibility among different 3D GIS systems.
After years of research in GIS (Geographic Information System), people have accumulated a number of ways to solve the above problems [3]. However in virtual 3D environment, because of the difference in data organization and manners of human computer interaction (HCI), we need to apply a new standardized method of modeling and scheduling for 3D models. At present, there is no such a uniform method as the constructing specification or standard for the modeling of 3D buildings. Existing approaches are insufficient and inefficient in the scheduling of large-scale building models, resulting in poor performance or large memory occupancy. In response to such questions, the authors proposed a new method for the construction of 3D building models. Models built using the proposed methods could be much better than those built using traditional methods. For the 3D modeling of building objects in scenes of large scale, the proposed methods can not only remarkably reduce the complexity and amount of model data, but can also improving the reusability and factuality of models. Concerning the scheduling of large-scale building models, the Model Loading Judgment Algorithm (MLJA) proposed in this paper could solve the optimal judgment problem of model loading in 3D vision cone, particularly in circumstance with uncertain user interactions.
This paper first examines and analyzes existing problems in constructing and scheduling steps of 3D building models. Then the authors propose a set of constructing rules for 3D building models together with methods of model optimization. Besides, special scheduling technology and optimization method for model rendering is also applied in this paper for large-scale 3D building models. In order to evaluate the efficiency of proposed rules and methods, a case study is undertaken by constructing a 3D model for the main campus of Peking University and Shenzhen using both the proposed method and the traditional method respectively. The two resulting 3D models of Peking University campus and Shenzhen are then examined and compared with one other in aspects of model data volume, model factuality, speed of model loading, average responding time during visualization, compatibility and reusability in various 3D geo-visualization platforms like China Star (one China’s own platform for 3D global GIS manufactured by the authors), Skyline, etc. Result of comparison tells that provided similar factuality of models, using the proposed method of us, the data volume of models was reduced by 86%; the speed of model loading was increased by 70%; the average responding time of model during visualization and interaction speed was reduced by 83%. Meanwhile, the compatibility and reusability of 3D model data are also improved if they are constructed using our approach.
II. MODELING RULES OF 3D BUILDINGS
3D scene is the best form of visualization for digital city systems. While constructing 3D models for buildings objects, proper methods and rules should be used, which are made with full concerns of the characteristics of 3D building models [2]. The resulting models should be robust, reusable and suitable enough for transmission over computer network, and should at the same time be automatically adapted to system capability.
Generally speaking, methods of constructing 3D building models can be classified into three types: wireframe modeling, surface modeling and solid modeling. In normal circumstances, to model buildings in 3D format, the framework of building should be constructed first according to the contour features, number of floors, floor height, aerial photograph and liveaction photos of buildings. Then, gather the characteristics of scene that the buildings to model are representing. Important characteristics include buildings aerial photograph or liveaction shooting photos. Finally, map the gathered texture to model framework, optimize the model and create database of the 3D building models.
Although there have already been many approaches for the construction of 3D building models, a unified modeling method and rules are still needed to improve the efficiency, quality, facilitate checking, reusability and archiving of constructed models. By investigating the characteristics of buildings, we found that buildings have regular geometric solid for modeling, similar texture on the surfaces of different directions, high similarity in small-scale models of buildings, etc. According to these, this article gives a discussion on the modeling rules from three aspects, including constructing rules of the 3D building models, texture mapping rules of 3D building models and optimization method for constructed models based on mentioned constructing rules.
A. Constructing rules of the 3D building models
The 3D building modeling refers to the procedure of representing true buildings from the real world into computer in the form of 3D objects [4]. Human beings, as the creator and at the same time potential users of models, play a key role in this procedure. People are different from each other in the understanding of the building objects, methods of modeling and the software tools they use for modeling. Such differences among people who carry out modeling work at the same time lead to the 3D models of diverse quality and low efficiency. So the 3D building constructing rules proposed in this article become necessary and helpful to solve the above problems.
1) Combine similar floors as a whole and keep the roof independent
2) Share similar models and process the details especially
3) Constructing in the unit of meters
4) Define central point of the model
5) Unified model codes
6) Reduce number of surfaces in a single model
7) Reduce combination of the models
8) Rational split of models
B. Texture mapping rules of 3D buildings
Based on the framework of 3D models, we need to attach these models with proper textures to create a better visualization effect for 3D buildings. The quality of texture mapping has a direct impact on the visual effect of the scene whiling being rendered [5]. Since the graphics card of computer will load all the textures together when rendering a model, texture mapping rules and the quality of the texture mapping can directly influence the efficiency of rendering as well.
C. Optimization of models based on constructing rules
Based on constructing rules and the characteristics of 3D building models, the authors develop a software tool to optimize the 3D building models automatically. The optimizations implemented in the software tool contain the deletion of models’ internal textures, merging adjacent vertices/lines/surfaces, removing un-mapped framework and so on. Besides, the software can enhance the shape of the whole model, texture position and model facticity in the procedure of model optimization.
III. SCHEDULING TECHNOLOGY OF LARGE-SCALE 3D BUILDING MODELS
For the 3D visualization of large-scale architectural models, a series of measures could be applied to ensure the efficient rendering of models. Important measures includes the scene organization, vision cone cutting, elimination of textures on the backside of models, Shader optimization, LOD Algorithm, math library optimization, memory allocation optimization, etc..
How to display thousands of 3D city buildings’ models in a virtual 3D environment is not trivial. The main problem is the scheduling of models [7]. It determines when and which models to be loaded. This problem can be divided into two smaller problems: Find visible spatial region of models in 3D environment, and optimization method of model rendering efficiency.
A. Find visible spatial region of models in 3D environment
According to operating mechanism of computers during 3D visualization and the characteristics of large-scale 3D scene, we need to determine the position of current viewpoint first before loading signal models or urban-unit models. Then in response to the regular changes of viewpoint in virtual 3D environment, the system will preload the 3D model data into memory automatically. In this way, frequent IO operations can be reduced and thus overall efficiency of system gets improved. A new algorithm named MLJA (Model Loading Judgment Algorithm) is proposed in this paper in order to find out visible region of models in the 3D environment. The algorithm integrates the graticules and elevation information to determine the current viewpoint of users in the 3D space. And with the movement of viewpoint, the algorithm schedules the loading of model correspondingly and efficiently.
B. Optimization method of model rendering efficiency
The scheduling method of large-scale 3D building models proposed above is an effective way to solve the problem caused the contradiction between large model data volume and limited capacity of computers. According to the algorithm, we can avoid loading the whole large-scale 3D building models at one time for the sake of limited computer memory, and then improve system efficiency in the procedure of model loading and abandoning. Due to the limited capacity of GPU and local video memory, we need a further research on how to display the loaded model data in more efficient manner. In the remaining part of this paper, the authors will continue to introduce several methods on the optimization of model rendering in the vision cone.
1) Elimination of textures on the backside of models
The backside of the 3D model is invisible to the users. If we omit the texture mapping for the 3D model on the backside, the processing load of graphic card will be reduced as much as at least 50%. Besides, according to an investigation on procedure of actual model rendering, the authors found that on the backside of the 3D model, the invisible texture is rendered in a counter-clockwise manner against the direction of eyesight, while the visible texture mapping is rendered in clockwise manner. So we can omit the rendering of models which is intended to be rendered in counterclockwise manner. Therefore, the textures won’t exist on the back of 3D models. The graphic card could then work more rapidly and efficiently.
2) Eliminate the shielde