Geoscience Reference
In-Depth Information
3 Indoor Dataset
The algorithms developed require to be thoroughly tested in an extensive and
complex indoor environment to be a valid alternative for outdoor algorithmic
testing. Although the authors realize that using a single specific building data-
set for testing can still be too limited to generalize the obtained results, we tried
to map a building with several features that are quite common for many indoor
environments. The dataset for our tests consist of the 'Plateau-Rozier' building
of Ghent University. It is a complex multi-storey building where several wings
and sections have different floor levels and are not immediately accessible. It is
assumed that the mapped indoor space is complex enough with many corners
and decision points to assume reasonable wayfinding needs for unfamiliar users.
Previous research executed in this building has shown that unfamiliar users can
have considerate difficulty recreating a previously shown route through the build-
ing (Viaene et al.
2014
).
The dataset is based on CAD floor plans which are transformed to ArcGIS
shapefiles for additional editing and querying. For application of the least risk
and shortest path algorithm, the original floor plans are converted into a three-
dimensional indoor network structure. Automatic derivation of indoor networks
has long been focused on as one of the problematic areas for indoor navigation
applications. Recent efforts have shown possibilities of automatically assigning
nodes to each room object and connecting them when they are connected in reality
(Anagnostopoulos et al.
2005
; Meijers et al.
2005
; Stoffel et al.
2008
). However,
the development of a comprehensive methodology for automatic network crea-
tion requires a thorough foundation and agreement on the appropriate and optimal
(i.e. user friendly) network structure of indoor environments which supports the
user in his navigation task (Becker et al.
2009
). Therefore, in most existing indoor
navigation applications, the data is still mostly manually transformed into graph
structures. As such, we decided to manually create the network based on the subdi-
vision into separate rooms (Fig.
3
).
The network structure is chosen to be compliant to Lee's Geometric Network
Model (Lee
2004
) as this is one of the main accepted indoor data structures. In this
model, each room is transformed into a node, forming a topologically sound con-
nectivity model. Afterwards, this network is transformed into a geometric model
by creating a subgraph for linear phenomena (e.g. corridors), as such enabling net-
work analysis. The position of the node within the rooms is chosen to be the geo-
metrical centre point of the polygons defining the rooms. This premise implies that
the actual walking pattern will sometimes not be conform to the connectivity rela-
tionships in the network inducing small errors in the calculations of shortest and
least risk paths. We will need to verify whether or not this error is significant in
the total cost of certain paths. The selection of corridors to be transformed into lin-
ear features is based on the map text labels indicating corridor functionality. These
areas also appear to be perceived as corridors when inspecting the building struc-
ture itself in the field. Obviously, this topic is depending on personal interpretation
