PERCEPTION-DRIVEN SIMPLIFICATION METHODOLOGY OF 
3D COMPLEX BUILDING MODELS* * 
Zhiqiang Du**, Qing Zhu, Junqiao Zhao 
State Key Lab of Information Engineering in Surveying Mapping and Remote Sensing, Wuhan University, 129 LuoYu 
Road, Wuhan, Hubei, 430079, P. R. China - (duzhiqiang, zhuqing, zhaojunqiao)@lmars.whu.edu.cn 
Commission V, WG V/4 
KEY WORDS: LOD Modeling, Complex Building Models, Human Visual Perception, Wavelet Transformation, Simplification 
ABSTRACT: 
The 3D complex building models are commonly of massive data, detailed appearance and sophisticated topology. The LOD 
modelling and real-time visualization of such models have hampered the wider applications of Virtual Geographic Environments. 
Aimed at the drawback of traditional simplification methods—difficult to locate the components, primitives that have to be 
decimated, this paper proposes a novel simplification method to locate the model portions need to be simplified based on the 
rendered image analysis. The main idea is to convert the recognition and assessment of the complex object’s geometric features in 
3D object space to the 2D image space, and to use the image filtering based on Human Visual System (HVS) model to guide the 
simplification operations in 3D space. Taking a complex timber-frame building model as an example, the experiment suggests that 
the method can effectively reduce the imperceptible details on the model surface, so that the data volume is decreased dramatically 
and the model’s shape is hold well. 
1. INTRODUCTION 
1.1 Motivation and Aims 
With the development of geospatial information technology, the 
age for wide use of 3D geographic information system (3D GIS) 
is coming up (Butler, 2006). As the core technique to support 
3D Virtual Geographic Environments (VGEs), 3D GIS should 
offer the most foundational spatial data processing functions 
such as data acquiring, arrangement and management, data 
analysis and visualization presentation. 
In the process to build VGEs, the fidelity requirements of 
different city models are increasing dramatically. Better 
incorporate presentation for inner and outer building model is 
desired, so more attentions are paid on constructing integrated 
inner building 3D model besides its surface reality 
representation, which benefits a lot in fields like virtual cultural 
heritage protection, 3D cadastre management and city disaster 
prevention. As a way of abstract presentation for real buildings, 
the block models built up based on different measuring means 
is far behind various requirements, especially for the high 
fidelity in large-scale Virtual Architecture Environments 
(VAEs). Thus, the modelling means based on CAD and LIDAR 
techniques are widely used in real 3D information acquiring 
and representing for complex buildings. By the introduction of 
these two modelling ways, a sharp increasing amount of model 
data has caused tremendous difficulties for smooth real-time 
roaming in various virtual environments. 
The complex building models mainly acquired and built by 
LIDAR or CAD techniques have highly detailed appearance 
and complicated inner structure, highly complex topological 
relationship inside and complex semantic relationship among 
the parts. And this kind of model is made up of various 
independent parts owning different types and configurations. A 
single building model, for instance, has such a huge data 
amount as millions of triangles. 
Now, the bottle-neck problem that restricts the wide use of 3D 
virtual geographic environment is the multi-detail level 
expression for those various natural and human high-precision 
3D models. But, the complexity of 3D models seems to grow 
faster than the ability of the hardware to render them. Thus no 
matter how to accelerate the tremendous improvement of 
computer graphic hardware, LOD modelling technique is also 
tightly correlated with the conflicts between the fidelity of 
protracting effect and the rendering speed. And because the 
higher detail the object has, the bigger its data amount would be, 
the more elaborate it can describe, vice versa. The fundamental 
aim of LOD modelling is to realize to express the most complex 
target by using the smallest data volume and to render the 
fidelity. 
1.2 Related Works 
The transformation form the highest detail model to the lowest 
is mainly by the way of model simplification. In order to 
automatically create LOD models, researchers have brought 
forward many kinds of simplification algorithm for polygon 
mesh in recent 30 years, and then the automatic creating LOD 
model is transformed into simplification of 3D polygon mesh, 
the common way in it is to decimate the unimportant primitive 
(e.g. vertices, edges and triangles) from the polygon mesh. The 
methods work well on the simplification for consecutive free 
curved surfaces such as terrain surface denoted by grid/TIN 
The work described in this paper is supported by the National Natural Science Foundation of China (40671158), the National High 
Technology Research (2006AA12Z224) and Development Program of China (2004CB318206). 
* Corresponding author.