The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B3b. Beijing 2008 
738 
S=S- 
ÔS „ , sin«' 
— -ah- h 
dh tan A' 
(x,y)- 
DEM 
->hu 
(10) 
(11) 
The terms dl / dh and ds / dh can be derived from RPC (7). 
They can be considered as local satellite angles. We assume 
that the base of a building lies on the ground with height h b 
(11) of DEM (SRTM). 
iii. a custom designed 2-D virtual memory management 
system for the optimal performance balance between 
image rendering speed and hardware requirements; 
iv. a user-friendly interface for collecting outlines of new 
models. 
The visualization platform provides user 2D and 3D view of the 
objects simultaneously. 
5.1 Hierarchical Level of Detail 
The method using RPC to determine the base or shadow of the 
top from changing the height is obviously more accurate than 
the conventional methods which locate the base and shadow 
first, measure the distance, then derive the height of an object. 
The new method reduces adjustable variable to height only so 
as to improve the accuracy. In our work to follow, we will 
concentrate on using the new approach to extract the building 
heights. The conventional methods can be use if RPC is not 
available. 
In a satellite image, it is common that the base of a building is 
obscure by the surrounding structures, or the shadow may lie on 
top of the nearby building or be occluded by the building itself 
or nearby buildings. Depending on the situation, the new 
approach provides users the choices simultaneously. Figure 3 
shows an example of the new approach to obtain the correct 
height of the building when only shadow can be observed 
clearly. 
5. SYSTEM DESIGN 
The system is based on our previous in-house development for 
a 3D city model visualization system (Huang et al., 2006). 
Figure 4 shows the system design. Briefly it implements two 
types of tasks, one is to visualize the existing 3D models in 
realtime, second is to reconstruct the new models 
simultaneously. In order to render large amount (Gigabytes) of 
data of satellite image, thousands of 3D models, a system has 
been developed to support: 
i. an efficient database for managing DEM, satellite level 
2A image and ortho-rectified ground image; 
ii. a level of detail (LOD) method for reducing 
computational load; 
3D Visualization 
In the system, the ortho-rectified image of the original IKONOS 
level 2 image is used as ground texture in the 3D viewer. It is 
generated when the level 2 image is ortho-rectified with the 
refined RPC and the same DEM. The level of detail (LOD) to 
DEM for terrain, to ortho-rectified image for ground, and to 
level 2 image for 3D feature is applied. The automated LOD 
selection will be performed by the engine of the platform 
according to the distance between the camera’s view point and 
the target. 
5.2 Software User Interface for 3D Extraction 
The software user interface is shown in fig. 5. The overview of 
the image coverage is shown on the upper-left comer of the 
window, the 2D satellite image is shown on the right of the 
window, while the 3D models, imported and newly generated, 
are shown on the lower-left comer of window, the 
orthorectified image and DEM are shown on the ground. Users 
are free to switch between 2D and 3D view as full screen. The 
software provides user a set of tools for various uses. The 
drawing tools are used for digitizing the building top in 2D 
view. The guidelines from the top to base, and from the top to 
shadow are drawn automatically according to the RPC, sun 
geometry and user candidate building height. The candidate 
building height can be either input from the keyboard, or varied 
by rotating the mouse central wheel. 
An example of the software application is shown in fig. 6 and 
fig. 7. Figure 6b and 7b are part of the same satellite image. The 
outline of a building top has been drawn in yellow by the user. 
The guidelines in cyan are from the building top to base, and 
Figure4. System Design 
Figure5. Software user interface