In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009 
datasets transfers, especially with slow home or mobile modems. 
Level of details improves the rendering speed but does not 
affect the download time, and on the other hand, the raw 3D 
scene model data imply severe drawback for data security and 
copyright issues. So we transmit only video sequences but no 
raw data to the mobile clients. After the 3D scenes data 
transferred to the mobile client, the frame rate was sufficient for 
conducting the navigation test on mobile client. 
In the experiment which measured the difference between 
simplify scene and primal scene, we recorded the amount of 
primary triangles transmit to client memory is about 70,000 (see 
figure 7a), and after being simplified, the amount of triangles 
reduced to 16064, almost 77.1 percent triangles are removed or 
cached (seeing figure 7b). 
a b 
Figue7. Comparison between primary scenes and 
simplified scenes, a) Primary terrain scenes, b) 
Simplified terrain scenes 
In the experiment which test the efficiency of 3D scenes 
generating and rendering, the test area we selected spans about 
200KmX200Km, and it was covered by 30.0m resolution ETM 
color photography and some attention areas were covered by 
1.0m resolution color aerial photography (about 3.5 GB), as 
well as 120 MB of DEMs which the highest resolution is 16m 
spacing and the lowest resolution is 256m spacing. On Pc server, 
we recorded the rendering efficiency about 80-120 fps with 
50%-60% CPU utilization and a delay inserted between the 
frames to maintain constant frame rates. 
a. b. 
Figure8. Panorama view of large terrain view with 
30.0m resolution ETM color image (a) and city view 
with 1.0m resolution color aerial image (b) on 
mobile client 
triangles and the facade texture of one building is about several 
million bytes. The average frame rate on servers was about 30- 
50 fps for the original texture, and the textures were DDS 
compressed and using mipmap textures, the average frame 
speed up to 70-100fps during real-time rendering. 
b. 
Figue9. Panorama of the city models, a) City 
models on Pc servers, b) Sketch the city 
modek on mnhile client 
6.2 Conclusions 
Mobile devices currently have the capability to request and 
display 3D panorama scene. This paper proposed an approach 
to generate 3D environment visualization system for personal 
navigational purposes that handles large heterogeneous datasets 
at multi-resolutions. The attached GPS provides the location 
information, and network servers provide the data and 
visualization processing. In the stage of offline data 
preparations, the full area is divided into regular tiles and a 
pyramid mode for multi-resolution virtual environment is 
generated. Having the virtual environment being divided into 
zones helps the users to narrow down their search towards or 
within the intended zone or category only, and it only need 
transmit compressed imagery that is actually requested by the 
users. By the client/server mode, the presented approach allows 
mobile applications to provide users interactive access to 
complex 3D scene models including high-resolution 3D terrain 
geometry, 3D building geometry, and textures. In particular, the 
server can be optimized for processing large-scale 3D scene 
models using high-end computer graphic hardware, whereas on 
mobile clients, there only multimedia capabilities are required. 
6.3 Future work 
In the city model experiment, over five hundred building 
models were rendered, and each building contains hundreds of 
The major problem with detailed 3D scene models is the big 
size, which affects both the rendering speed and the download 
113