ul 2004 
- Berlin: 
ples of 
lodeling 
tions of 
national 
Sweden. 
AUTOMATIC GENERALIZATION OF ROADS AND BUILDINGS 
Pingtao Wang, Takeshi Doihara 
Asia Air Survey Co. Ltd. 
Kanagawa 215-0004, Japan 
{pt.wang, t.doihara }@ajiko.co.jp 
Commission VI, WG IV/3 
KEY WORDS: Generalization, Automation, Centerline, Buildings, Block, Road, Cartography. 
ABSTRACT: 
Map generalization simplifies the details of map representation. Automatic generalization has been a hot research topic for decades, 
but there does not exist a set of universal rules or algorithms that explicitly defines how generalization should be performed. This 
paper presents a method to automatically generalize roads and buildings. With the proposed method, road generalization and 
building generalization are carried out consecutively. Road generalization includes Road Modeler, which converts original road 
edges to road polygons, and Network Generator, which collapses road polygons to road networks. The created network is 
topologically connected and suitable for GIS (Geographical Information System), such as car navigation systems. Building 
generalization is mainly composed of clustering building polygons to building clusters, aggregating a building cluster to a polygon, 
and simplifying both original and aggregated building polygons. Using the created road networks as the constraints for generalizing 
buildings leads to the generalized results without contradiction. Some experiments have also been implemented to verify the 
effectiveness of the proposed method. 
1. INTRODUCTION 
A map is a resolution-dependant geographical representation of 
the real world. Map generalization is a complicated process and 
usually involves a great deal of spatial analysis to decide what 
and how to generalize, and how to resolve conflicts that might 
occur during the process. Manual generalization is a time- 
consuming and skilled work. Cartographers draw a reduced map 
by hand, and eliminated some unimportant features to simplify 
lines, to combine adjacent areas, and to resolve conflicts as their 
judgment (ESRI, 1996). That is to say, manual generalization 
leads to the inconsistent results because of the difference of 
cartographer's experiences. Therefore, automated map 
generalization is desirable and has been researched for decades. 
Nowadays, with the development of computer science, some 
manual processes of map generalization are being implemented 
on some GIS software. There are many researches about 
automation of generalization operators such as simplification, 
displacement, collapse, aggregation, typification, and so on. 
Lines are major features in a map, and most efforts about 
automatic generalization have been made in  line's 
generalization or simplification. Furthermore, roads and 
buildings, which are usually represented by lines, are the most 
basic objects in a digital map. Therefore, we shall pay special 
attention to the automatic generalization of road and building 
data in this paper. 
Road generalization is the process of creating and/or updating 
the road network of a small-scale map from the corresponding 
road edges of a large-scale map. Line simplification, which may 
be the earliest attempt for automating generalizafion (Douglas, 
1973), has been widely used to simplify road networks (lines) 
and other linear features. Kreveld and Peschier presented an 
approach to generalize road networks by keeping three 
objectives in mind: not allowing roads to be too close, avoiding 
249 
detours between important points, and giving priority to bigger 
roads (Kreveld 1998). Annita and others tried to collapse 
polygonal road area to road network by using triangulation 
(Annita 1998). There are also many other papers about the 
processing of road centerlines or generalization of polygonal 
roads, but rare researches are found for creating road network 
from road edges or double-line roads. Considering the fact that 
roads are widely represented by double-lines, rather than 
polygons, in large-scale maps for our experiments, further 
researches are necessary to convert road edges to road networks 
directly. Here, double-line roads may be the curb lines or the 
boundaries of the corresponding road area in a large-scale map. 
Building generalization involves the simplification of 
independent building polygons, the aggregation of building 
clusters and the displacement between the generalized buildings 
and other features such as roads. For building cluster 
aggregation, Regnauld developed a method to detect building 
pattern groups by applying the minimum spanning tree (MST) 
model from graph theory (Regnauld, 1996). Anders and Sester 
applied hierarchical cluster algorithm to typify buildings and 
lakes (Anders, 2000). From the legibility of the entire 
generalized map, some other objects, such as roads, should also 
be considered in building generalization. 
In this paper, a new framework is proposed to generalize road 
and building data. The present framework implements road 
generalization and building generalization consecutively. Road 
generalization includes Road Modeler, which converts original 
road edges to road polygons, and Network Generator, which 
collapses road polygons to road networks. The created network 
is topologically connected and suitable for GIS (Geographical 
Information System), such as car navigation systems. Building 
generalization is mainly composed of clustering building 
polygons to building clusters, aggregating a building cluster to a 
polygon, and simplifying both original and aggregated building