International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
1.1 Location aware applications
If we have a closer look on the market we will notice that there
are several simple information systems providing information
dependent on the actual location. As these commercial systems
are offered by different companies each system uses a
proprietary data format. Access on data of different information
providers and exchange are difficult.
An open platform for location aware applications can widen the
possibilities, as everybody would be able to contribute
information to the model. Therefore one challenge of the
research within the Stuttgart University project NEXUS is the
provision of such an open platform — the NEXUS system. This
system relies on a model-based concept, called the NEXUS
Augmented World Model (AWM) (Nicklas et al., 2001). The
AWM is the base for the NEXUS system’s extensibility and
flexibility and it forms the interface to the applications. As it is
an open platform also existing data sources like the WWW shall
be integrated. This may lead to a great heterogeneity in the data.
In NEXUS a federation approach is used to handle that
heterogeneity, see Fig. 1.
A closer view to the object oriented AWM shows its basic idea:
federation of information and representation of the real world
(buildings, streets, cars,...). As example one representation of
the real world could be a detailed 3D city model. For interaction
with the AWM and the use of NEXUS services the architecture
provides also an interface for sensor integration. Here different
positioning sensors can be plugged in to provide the necessary
position and orientation information to several applications.
|
“|| APPLICATIONS E
temen | DARREN
i € A
i * ne
Navigation ... ,Smart Factory City guide
sensor s
data Node 1 Noden
Augmented World Model
digital libraries
Figurel. Architecture of the NEXUS platform
2. LOCATION SENSING
2.1 Techniques for location sensing
Position information is the fundamental requirement of location
aware applications. To provide this information different
techniques can be used. In principle there are three methods for
automatic location sensing: triangulation, proximity and scene
analysis (Hightower & Borriello, 2001). Triangulation
techniques use geometric properties of triangles to compute
object locations. This method is divisible into the subcategories
of lateration, using distance measurements, and angulation,
using primarily angle measurements. As example systems like
GPS or a magnetic compass use triangulation techniques. When
an object is “near” a known location then this is described with
904
proximity. For sensing proximity there are three general
approaches: (a) Detecting physical contact, (b) monitoring
wireless cellular access points and (d) observing automatic ID
systems.
A further method for location sensing is the scene analysis
technique. Here features of an observed scene are used to
conclude to the location of the observer or to the location of
objects within the scene. Usually the observed scenes are
simplified to obtain features that are easy to represent and
compare.
The position and orientation of a user are basic information to
provide high quality location based services. Azuma (Azuma et
al., 1999) evaluated that all tracking systems lack accuracy or
robustness. This leads to his conclusion that only a combination
of different technologies, which he call a hybrid tracking
system, should be used. Several systems are using GPS and
additional devices to track the orientation. But often the existing
approaches are not accurate enough for the overlay of a
reconstruction of real world objects or they cannot be applied to
persons walking in a city without the requirement of fixed
positions where they should stand to receive information. The
Touring machine of Columbia University in New York is one of
the most well known information systems (Julier et al., 2000). It
broadcasts information about names of. the buildings, so
positioning and orientation accuracy are not the most important
conditions. To provide information about more specific
building features the requirements on position and orientation
accuracy are higher. The use of image processing techniques is
a method to improve results. Beveridge (Beveridge et al., 1996)
and Behringer (Behringer, 1999) used horizon shapes extracted
from a visual scene to look up the observer's location from a
prebuilt dataset. In You (You et al., 1999) an augmented reality
system is described which tries to correct drift errors of a
gyroscope and errors of the compass using images collected by
an additional camera.
Pedestrians usually use viewpoint information and landmarks
(e.g. buildings) to locate themselves in a familiar environment.
To facilitate the same in an unfamiliar environment an
automation of this process should be provided by combining the
model of the environment and surrounding objects (landmarks).
In the research project NEXUS a model of the real world exists —
the Augmented World Model. This model contains various
information as well as a 3D representation of real world objects
(e.g. buildings). On that condition we are able to integrate
divers sensors and operations on these data to support
pedestrian orientation and navigation.
2.2 Orientation and navigation using scene analysis
In the last years the availability of low-cost imaging devices
increased. The combination of mobile computational
capabilities, imaging capabilities, positioning devices and
network access opens a door for novel applications. As example
Augmented Reality applications in urban regions are useful to
assist users and to interact with the model of the environment.
Especially in NEXUS where a model of the environment exists
information about objects can be provided by identifying them.
In order to allow information access or to use objects as
landmarks for navigation, the link between the Augmented
World Model and the observed environment has to be
generated. As we concentrate on image information this leads to
an approach which tries to overlay the model and the
corresponding primitives in the real world. Similar approaches
often use a database where manually geo-referenced sequential
images are registered. Then through the registered image the
landmark lines are transferred on the other unregistered images
