ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS’’, Bangkok, May 23-25, 2001 
The result is a list of co-ordinates structured according to the 
order of the faces and order of the nodes in a face. To create the 
VRML file, the data set obtained with these queries has to be 
further structured according the VRML model. In our example, 
two options are possible to represent the faces constructing the 
POLYHEDRON 23 in VRML, i.e. they can be stored as individual 
faces or as a part of one polyhedron. The first option is simpler 
and can be derived directly from the SQL query. The second 
representation requires control of duplicated co-ordinates and 
the correct ordering of the corresponding co-ordinates in the 
description of the faces (i.e. in the VRML node coordlndex). 
Currently, we concentrated on development of operators that 
create a VRML file consisting of objects (i.e. not individual 
faces). Since standard SQL statements cannot perform this 
operator, the computations have to be completed with the help 
of a host language (PL/SQL, C++, Java). At this stage, we have 
used PL/SQL, i.e. the script language provided by Oracle. 
A number of tests aimed at clarification of the best mapping and 
fastest operators to retrieve the needed data. For the purpose 
the first tree mappings discussed in the paper are implemented 
and populated with data. The corresponding VRML creators are 
tested for performance. Although the data structure is at very 
initial stage, the first results are very encouraging: 20 000 faces 
(part of 1600 buildings) can be retrieved in less then 23 seconds 
for representation with nested tables and less then a second for 
the relational representation and object-oriented views. The 
conditions of the test and the data sets are reported and 
discussed in more details in [17]. In principal, we expect less 
data to be extracted for rendering, i.e. the objects that are visible 
only from the current position of the user. These data are 
estimated in the range between 50 and 5000 faces if spatial 
search is applied. This is indication for a possible further 
improvement of the obtained results. 
7 CONCLUSIONS 
We have presented a 3D topological structure that provides data 
for a real time application i.e. serves two tasks (pose 
determination and rendering of virtual objects) that require real 
time commuting. The proposed structure maintains four 
abstractions of geometric representation (the ones mostly 
employed in 3D modelling) based on two constructive elements 
(faces and nodes). To be able to provide “cheap” details in terms 
of line features, the model incorporates non-topologically 
organised data, i.e. lines. Explicit relationship links a set of lines 
to a face. To be able to achieve the required accuracy and to 
build the 3D topology, a number of 3D reconstruction methods 
are applied. The model is implemented in relational database 
Oracle utilising relational and object-relational mappings. 
Several operators to create a VRML file are created and tested. 
The experiments clearly show that a 3D topological model can 
be adopted for an augmented reality application. The 
performance of the mappings in relational database drops 
bellow the required 6 seconds. The methods utilised in the 3D 
reconstruction ensure accuracy of few decimetres that is agreed 
to be sufficient for the positioning system. Therefore we consider 
the results reported in this paper a successful step toward a 3D 
GIS supplying data for a real time application. 
Still more experiments are needed to clarify the relational 
mapping that will assure the best performance. Currently, the 
SQL queries are executed from the Oracle high-level language 
that cannot be integrated in the UbiCom architecture, i.e. C++ 
modules have to be developed and further tested. Location and 
efficient spatial search in such large databases can not be 
performed without appropriate spatial indexing. One of the 
directions for further research within the project is related to 
developing a set of specific operations than will reduce the 
amount of data transmitted to the vision system. Examples of 
such operators are determination of the area of interest (using 
approximate positioning obtained by the mobile equipment), 
back-face culling (to eliminate invisible faces) and a variety of 
line filters for retrieval of line features. 
References 
[1] Abiteboul, S and R. Hull, 1987, IFO: A formal semantic 
database model, ACM Transactions on Database Systems, 
Vol.12, No.4, pp. 525-565 
[2] Bodum, L., I. Afman and J. Smith, 1998, Spatial Planning 
moves out of the flatlands, AGILE, 23-25 April, Enschede, 
The Netherlands, CD-ROM 
[3] de la Losa, A. and B. Cervelle, 1999, 3D topological 
modelling and visualisation for 3D GIS, Computers 
Graphics, Vol.23 
[4] Kofler, M. and M. Gruber, 1997, Toward a 3D GIS 
Database, GIM, Vol. 11, No. 5, pp. 55-57 
[5] Molenaar, M., 1990 A Formal Data Structure for 3D Vector 
Maps, Proceedings of EGIS’90, Vol. 2, Amsterdam, The 
Netherlands, pp. 770-781 
[6] OpenGIS specifications, 2000, available on 
http:. ;/ www.openqis.orq/techno/'specs-htm 
[7] Pasman, W., A. van der Schaaf, R.L Lagendijk and F.W. 
Jansen, 1999, Low latency rendering for mobile augmented 
reality, Proceedings of the ASCI'99, 15-17 June, Heijen, the 
Netherlands, pp. 372-376 
[8] Persa, S and P. Jonker, 1999, On Positioning Systems for 
Augmented Reality Applications, Handheld and Ubiquitous 
Computing 1999, Springer Lecture Notes in Computer 
Science 1707 
[9] Pilouk, M., 1996, Integrated Modelling for 3D GIS, PhD 
thesis, ITC, The Netherlands, 200 p. 
[10] Pigot, S., 1995, A Topological Model for a 3-Dimensional 
Spatial Information System, PhD thesis, University of 
Tasmania, Australia, 228 p. 
[11] Ubicom project, 2000, available on 
http://bscw.ubicom.tudelft.nl:' 
[12] van Oosterom, P., 1997, Maintaining consistent topology 
including historical data in a very large spatial database, 
Auto Carto 13, April, Seatle WA, pp. 
[13] Verbree, E., G. van Maren, R. Germs, F. Jansen and M.J. 
Kraak, 1999, Interaction in virtual world views—linking 3D 
GIS with VR, Int. J. Geographical Information Science, 
Vo!13, no 4, pp. 385-396 
[14] The Virtual Reality Modelling Language, 1997, available on 
http:7www.web3cl.orq. , technicaiinfo/specifications/vrinl97.-in 
dex.htm 
[15] Zlatanova, S., 2000, 3D GIS for urban development, PhD 
thesis, ITC public. 69, Enschede, The Netherlands, 222 p. 
[16] Zlatanova, S., J. Paintsil and K. Tempfli, 1998, 3D object 
reconstruction form aerial stereo images, Proceeding of the 
6th International Conference in Central Europe on 
Computer Graphics and Visualization'98, 9-13 February, 
Plzen, Czech Republic, pp. 472-478 
[17] Zlatanova, S. and E. Verbree, 2000, A 3D topological 
model for augmented reality, Proceedings of the Second 
international symposium on MMSA, 9-10 November, Delft, 
The Netherlands, pp. 19-26 
BIOGRAPHY 
Siyka Zlatanova: 
MSc degree at the University of Architecture, Civil Engineering 
and Geodesy, Sofia, Bulgaria in 1983 (Geodesy and 
Photogrammetry). PhD degree at the Graz University of 
Technology, Graz, Austria. Currently, a post-doctoral researcher 
at the Delft University of Technology, Delft, The Netherlands. 
Research field: Spatial Data Handling, 3D topology, 3D object 
reconstruction and visualisation.