You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

The 3rd ISPRS Workshop on Dynamic and Multi-Dimensional GIS & the 10th Annual Conference of CPGIS on Geoinformatics
Chen, Jun

1SPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001
Department of Geodesy, Delft University of Technology
Thijsseweg 11,2629 JA, Delft, The Netherlands
KEYWORDS 3D topology, 3D model, spatial objects, 3D visualisation, 3D GIS, real-time
This paper addresses an approach for storing 3D spatial objects with respect to the requirements of an outdoor augmented reality
application. The presented 3D data structure focuses spatial objects of urban areas, as the aim is support of 3D topology, high level of
details and appropriate performance. To satisfy these contradictory requirements, only the outlines of spatial objects are organised in a
3D topological model. Details on façades (e.g. doors, windows) are modelled as line features. An explicit relationship “line on face"
indicates the link between the lines and the corresponding façades. The paper concentrates on the 3D topological data structure, which
a typical example of so called boundary representations. The 3D data structure is implemented in Oracle 8/ DBMS using three different
approaches, i.e. relational, relational with object views and object-relational. The tests have showed that the proposed data structure
can be tuned to meet the needs of real-time rendering and positioning for augmented reality. We consider our results a promising step
toward extending 3D GISs to serve real-time applications.
Augmented reality is a mixture of reality and virtuality that allows
real world to be augmented with additional information. Virtual
objects (text or graphics) are visualised in the field of view via
special transparent glasses. The user can observe the
surrounding world and the virtual objects simultaneously (see
Figure 1). Until recently, the augmented reality application were
restricted to indoor applications e.g. surgery, inspection of
hazardous spaces. With the advances of the computer and
vision technology, augmented reality systems attempt to leave
the world of indoor applications, which rises new challenging
topics for research. Among them, structuring and database
organisation of the 3D model required for positioning and
visualisation of virtual objects is most pressing. Augmented
reality aiming at outdoor urban applications (e.g. rescue
operations, utility management, urban development, guided
navigation) needs a 3D model of size comparable to a town, i.e.
thousands of houses, streets, parking lots, etc. For example, the
national 2D topographic map of The Netherlands contains about
31 million line objects (see [12]). In residential areas this number
might increase three or four times for the corresponding 3D
model. Such an application faces all the problems in processing
and maintaining large 3D data sets.
This paper presents a 3D model aiming at both maintenance of
3D topology (one of the most important features of a 3D GIS)
and efficient organisation of 3D data for augmented reality
applications. The paper is organised in four sections: first the
requirements to the data structure are specified with respect to
the tasks of the vision system, second the proposed data
structure is discussed, then the 3D re-construction procedures
are briefly explained and finally some initial experiments within
Oracle database are reported. The research is a part of the
interdisciplinary UbiCom project carried out at Delft University of
Technology, The Netherlands (see [11]).
Figure 1: A person with mobile equipment and the observed
Discussions related to the content and the structuring of data in
3D GIS can be found in many publications on 3D GIS (see [2],
[3], [8], [13], [15]). Therefore, in this paper, we focus the specific
requirements to the 3D model with respect to the vision system
intended in the UbiCom project.
The 3D model is to be used for two critical subsystems of the
system architecture, i.e. positioning and visualisation (rendering)
of virtual objects. The expected types of data retrieved from the
database are different for both subsystems.
• Line features. The positioning system in UbiCom relies on
line features (straight lines) supplied by the database. The
term positioning refers to determining the accurate location
of the user (i.e. the person using the system) in the real
world. The movement of the user is followed (tracked) by
mobile equipment (GPS, accelerator and inertial system)
that provides an approximate positioning at range of 2-10
meters. The accurate positioning is to be achieved by
establishing the correspondence between lines extracted
from video images (provided by the video camera that is
also a part of the mobile unit) in real-time and lines
retrieved from the 3D model (see [8]). The success of the
line matching (and thus the accurate positioning) is closely
related to the amount of details maintained in the 3D
• 3D topology. In contract to positioning, the visualisation of
virtual objects requires no details but correct shape and
orientation of the real objects. Since the virtual object is
“mixed” with the real world, it is very likely a real object to
appear between the virtual one and the user. In this case
the real object is said to be an occluder of the virtual object
(see [7]). The user should not see the occluded parts of the
virtual object. In other words, the rendering engine has to
be able to compute which parts of the virtual object are not
visible in that particular moment and remove them from the
scene. The computations require the accurate location of
the user (provided by the positioning system) and the 3D
model of the real objects (retrieved from the 3D GIS). The
3D model has to ensure connectivity and continuity.
• Performance. The mobile equipment is capable of tracking
the movement in certain period of time without reference to
the database. This period depends on the speed of walking
and the movements of the head. The estimation is that
every 6-7 sec the system has to receive new data sets from
the database. Within this time (called lag of the system), a
query has to be sent to the database, processed and the
retrieved data has to be transmitted back to the mobile unit.
The performance of the database is one of the critical
issues influencing the lag of the system.
• Vector representation. Since both rendering and
positioning system use vector representation, we
concentrate on vector data structures.
. VRML. The Virtual Reality Modelling Language was chosen
as a standard to exchange data between the individual
subsystems. The language provides a full and flexible