Aulti-
Road
xt of
(4),
Delft
r und
piller
ns on
] and
Data
matic
rural
sition
puter
urnal
Fifth
mote
ction
1. In:
h. In:
chine
Pho-
'es of
267—
nsors
igery.
ty for
«shop
Shoichi Horiguchi
RECONSTRUCTING ROAD AND BLOCK FROM DEM IN URBAN AREA
Shoichi HORIGUCHI, Shiro OZAWA, Shigeru NAGAL, Kazuhiro SUGIYAMA
NTT, Japan
Cyber Space Laboratories
horiguti ? marsh.hil.ntt.co.jp
KEY WORDS: Urban objects, Reconstruction, Modeling, Texture mapping, DEM, Aerial images, GPS
ABSTRACT
This paper targets the reconstruction of 3D urban models consisting of realistic architecture models and different objects
on a ground surface. When reconstructing 3D urban models it is necessary to handle a diverse range of complex
structures and objects, and the models must be efficiently and precisely formed. Onto these models will be projected the
textures acquired from the actual objects. Research into building reconstruction has been active, but there is little
research on structures other than buildings. This paper describes a new approach to reconstructing models for roads,
intersections and blocks, that is to say, ground surface objects in urban areas.
When reconstructing road, intersection and block models there are three problems. The first is the separation of roads
and blocks. The second is the accurate determination of surface model parameters. The third is the construction of the
optimal model, that is, both the model degree and the model error is minimum Against the first problem, we use digital
2D maps and separate roads and blocks by matching the edge points of buildings acquired from DEM to the building
shapes on maps. Against the second problem, we extract only scattered elevation points to avoid obstacles such as
buildings, trees inside the block and cars on the roads, and determine the surface model parameters by analyzing the
partial cross sections of roads and blocks, using the MDL principle and heuristic construction knowledge.
Finally we show an example of 3D urban models of many buildings on a ground surface. Onto these models are
projected realistic textures.
1 INTRODUCTION
1.1 What's 3D urban model
3D urban model is needed in the fields of urban planning, urban investigation, and urban disaster simulations. Because
realistic texture is projected onto each model, the 3D urban model is also expected to be used more often in the
conservation of houses, reproduction of urban views, and forming the Digital City. Moreover, we expect to use the 3D
urban model as the infrastructure for the Geometric Information System (GIS) in urban areas. This system will build a
better urban life for us. Examples include traffic control, evacuation guidance, route guidance, and local information.
For putting these goals into practice we need detailed 3D urban model, especially road and block models, that is to say,
ground surface models are very important. They make it possible to simulate disaster scenarios such as inundation,
earthquake, fire, and traffic. By extracting polygonal ground surface models we can project realistic textures onto the
models to create walkthrough worlds.
1.2 State of Object Extraction from DEM
There are two main techniques that realize automatic object extraction for the recovery of 3D urban models. One
extracts buildings from multiple aerial images (Baillard, C. et al., 1999). The other extracts buildings from DEM
(Digital Elevation Map) data acquired by airborne laser scanning systems. In recent years laser scanning systems have
become a very attractive way to acquire 3D data. Especially when mounted in a helicopter, such systems can produce
high-density height data revealing detailed information about the presence and shape of buildings. By using a helicopter
we can efficiently and easily acquire 3D data. Compared to the height points determined by matching aerial images, the
airborne laser scanning systems yield very precise and reliable measurements. Therefore, we utilize DEM data acquired
by an airborne laser scanning system.
Several papers have been presented that deal with building extraction from DEM. The pioneers used parametric
building models and determined the shape and position parameters by fitting the models to the DEM (Haala, N., 1994).
They also reconstructed prismatic building models with flat roofs using the Minimum Description Length principle
(Haala, N., 1997). They then utilize the known ground plan information to improve model accuracy. The ground plan of
a building is subdivided in rectangles. Several models of building primitives are fitted to the height data within each
rectangle. Merging the models of the primitives using the results of best fitting forms the building models. Moreover,
they extracted planar roof faces from DEM. In this case, building plans were utilized to determine the orientation of the
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 413