CIPA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
In order to solve for the issues described in previous section, 
a VRML split-browser has been developed. Its working is 
based on a computer network model, where a central 
processor (the server) provides different kind of services to a 
number of hosts (clients). In practice, this kind of viewer is 
obtained by application of a client-server model to a generic 
VRML viewer. As shown in figure 2, the proposed model 
clearly distinguishes between server-side and client-side 
tasks. The server manages the whole virtual environment 
computing new scene views, processing user’s commands 
and executing scripts or animations. From the client-side, an 
interactive GUI is provided to the user by which the virtual 
views sent form the server are displayed. Moreover, the user 
can interact with the server by means of a number of 
navigation commands available on the client’s GUI. After an 
initial setup phase, needed to negotiate the parameters of the 
communication protocol (e.g. maximum throughput accepted 
by the client), a bidirectional communication loop begins 
between server and client. In practice, the user activates a 
command in the client’s GUI, through mouse and/or 
keyboard. Such command is transmitted to the server as 
event, i.e. as a request of scene change. Then the server 
intreprets the event and, if possible, sends back to the client 
the new view, which will displayed on the user’s GUI. The 
described loop is repeated according to the user interaction. 
^ 
T 7 
Graphical 
ùigme 
VRML 
Model 
I™ 
Figure2: server/client-based VRML viewer architecture 
This client-server approach shows following advantages: 
1) The amount of data sent to the client is greatly reduced 
respect with the transmission of the whole 3D model in 
VRML format. Indeed the server sends to the client an 
image sequence only: adopting an image compression 
technique, the throughput could be further reduced; 
2) The client doesn’t need high computational capabilities, 
as the world management step is fully performed on the 
server; 
3) Since server and client are independent each other, the 
LOD (Level of Detail) of each scene can be interactively 
changed by the client; 
4) The system features a certain degree of portability, as 
the source codes can be modified accordingly to the 
Operative System on which the server-side and the 
client-side applications should run. 
5) The copyright on the 3D model is keeped, as the user 
will never receive on his PC the whole VRML model, 
but only a set of images corresponding to the actual 
scene displayed on the monitor screen; 
6) The user’s GUI is managed by the client only, 
independently from the server. In this way a higher 
effectiveness of the whole system is ensured, since the 
server-side of the viewer doesn’t need to dedicate 
computing resources for the user’s GUI management. 
In order to allow the client to navigate the 3D model as fluent 
as possible, regardless the size and complexity of the model 
and client’s computing capabilities, above mentioned features 
of our proposed split-browser are still not sufficient. An 
image compression technique has to be implemented to get 
an optimized client/server communication protocol. To this 
aim two different compression algorithms have been 
investigated, the LZ77 and the JPEG, comparing the results 
of their applications to our split-browser transmitting simple 
geometric 3D shapes. However obtained results revealed a 
limited compression capability (higher for the LZ77 respect 
with the JPEG) for our goals. This can be explained 
considering that those algorithms are “general purpose”: they 
can be applied to whatever kind of file, regardless its content. 
Therefore we developed an “ad hoc” compression technique 
which would exploit the “knowledge” about the context 
where it should be applied. The description of this technique 
will be the topic of the following section. 
4. THE FRAME COMPRESSION ALGORITHM 
The alternative procedure we developed in order to further 
reduce the throughput between server and client is based on 
the application of projective transformations. The underlying 
idea is that subsequent views, processed by the server and 
displayed by the client to the user, are joined each other 
through such kind of transformations. Since the knowledge 
about the geometry of the virtual environment (e.g. eye-point 
position, viewing direction, etc) is needed, projective 
transformations don’t allow to preserve the same level of 
flexibility and versatility as LZ77 and JPEG image 
compression algorithms. Therefore our method can be 
applied to specific cases only, where geometry related images 
are handled. As shown in figure 3, the server/client communi 
cation procedure works as follows: 
Server-side operations 
1) Given the user input (scene change) at time n, the server 
processes the request generating the corresponding new 
view of the 3D model (view B). 
2) As the geometrical parameters describing the 3D scene 
at time n-I and n are known, the server can determine 
the projective transformation, mapping view A to view 
B. Computed parameters are then sent to the client. 
3) The view A, describing the 3D scene at time n-1, is 
applied the previous computed projective transforma 
tion. Consequently, a predicted view is generated, which 
should represent a good approximation ov view B. the 
prediction is generated by transforming the reference 
frame (view A) on the basis of the displacement of the 
user point of view (known from the selected command 
of the GUI) and the distance from the image pixel 
projection plane (known from the content of the Z 
buffer).