ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001 
84 
A pre-interpreted model matched a object with almost different 
heights; 
A pre-interpreted model can not match the object on the given 
range image; 
An object on the given range image can not match the pre 
interpreted object on the map. 
Figure7 Matching between models and range image objects 
Figure8 Object reconstruction by using the laser range 
In the first case, the matching result means that the object is a 
prismatic object and we can simply estimate the object height 
by using the average or the maximum height value within the 
matching region. In the second case, the match result means 
the object may be with the complicated structures. If we know 
some detail object features from vector maps, we can estimate 
these parameters by using the height values within the 
matching region. The last two cases of matching results mean 
that the object changes have been detected, in which the first 
case means that the old object has been deleted and the last 
case means that a new object has been created. For solving 
the change detected problems in our system now is based on 
semi-automated editing, i.e. the computer automated searching 
the changed places and operators modify these change places 
with manual editing. Figure 7 shows a basic matching 
procedure between map models and range image objects. 
If not only airborne laser range images but also ground laser 
range image in the different view points are available, above 
presented methods can be extended for reconstruction of high 
accurate 3D spatial objects with their detail features. The basic 
idea is to extract different DSMs from the different view points 
(as Figure 8). After that we can integrate these DSMs to 
generating whole 3D spatial objects by using the data fusion 
methods. In this case, the information got from 2D vector maps 
also can partially serve as a useful data source for estimating 
the vertical surfaces. Surface interpolation algorithms also 
should be extended for generating invisible object parts. 
3.4 3D visual Modeling 
There are several areas in urban system whose design and 
management can be considerably improved with the help of 2D 
or 3D visual modeling. Among these needs are traditional 
mapping, infrastructure design, urban planning and 
environment. Since there is not a clearly defined terminology 
for various types of 3D city models, we may simply call a 3D 
city model as a special computer representation of all fixed 3D 
spatial objects (buildings, vegetation, traffic- and waterways) 
within a urban area. 
For different purposes and applications, such as GIS related 
3D spatial management and analyzing, simulation and 
visualization of urban planning, and building design and 
construction, there are several kinds of 3D city models existed 
now. In these models the 3D spatial objects in urban areas are 
described as the different detail of their structures. Ranzinger 
and Gleixner have summarized four kinds of 3D city models 
mainly for simulation and visualization (Ranzinger and 
Gleixner, 1996.) The 3D spatial objects are called the prismatic 
model, in which the building is represented whither as a cubic 
box with the fixed parameters or as an object with the plain 
polygon adding the same height. The 3D spatial objects are 
called the parametric model, in which the building is 
represented with its roof structures. Other two kinds of 3D 
spatial objects show the complicated 3D object and with added 
surface 2D or 3D image textures. 
Our target is to generation of complicated 3D object and with 
added surface 2D image textures from air photos. By using 
Softlmage|3D system, we need generate the 3D spatial objects 
based on extracted spatial lines and surfaces. Here, the 
following kinds of 3D objects have been used for 3D city 
modeling in our system: 
• Polygon Mesh object: using for modeling grid DTM or 
surfaces 
• NURBS curves: using for modeling line objects 
• Patch and NURBS surface: using for modeling 
smoothing un-form surface objects 
• 3D TIN objects: using for modeling un-form surface 
objects 
• Boolean object: using for modeling and generating 
complicated objects 
3.5 Generation of Virtual Reality Environments 
One of main task for generation of Virtual Reality environment 
is to put the real image textures on generated 3D spatial 
objects. 
In order to use image as a texture of surface, objects will be 
taken as photographs in several parts and after that will be 
merged on surfaces (Dorffner, L.; Forked, G., 1998). However, 
the photographs, 2D texture-map, represented the objects 
might be mathematically transformed to the orthophoto, 
differential rectification, because the projection of photographs 
represent in central projection (Zhizhuo, 1990). At the other 
points within the segment, there still exist distortions due to 
relief displacements, and in the case of the indirect projection 
mode, distortions due to tilt displacements also exist. If we 
leave these displacements uncorrected, they will give rise to 
errors and hence effect the quality of the orthophoto (Gruen, 
1999). 
To generate computer models that are visually realistic, texture 
is applied to the reconstructed surfaces to make the models 
appear rich and complex. This be achieved by texturing the 3D 
triangulated surfaces from the laser scan with the appearance 
camera. Segments of 3D scene structure may appear in