Geography Reference
In-Depth Information
on the same region in a chain of steps or to investigate different events that
occurred at different time intervals or are overlapped.
For the sake of comparison with the existing approaches, we would like to
mention the model used for semantic enrichment of trajectories. Furthermore, for
the tasks in this paper that we regard as non-applicable, we will provide expla-
nations respectively. Most of the spatio-temporal data is obtained by GPS devices
and contains sequences of space-and-time referenced points measured at arbitrarily
chosen time intervals.
One of the widely used approaches in working with trajectories is based on
extraction of significant places from a single trajectory using an object's stay time
heuristic (Andrienko et al.
2007
). This approach was later conceptualized by
(Spaccapietra et al.
2008
) by introducing a model in which trajectories are divided
into sequences of stops (important places) and moves (movement to or from
important places). Two main approaches are used to find important places in
trajectories. The first considers only the characteristics of the trajectory (consid-
erable time spent in a place). In the second approach, important places are obtained
by intersecting the trajectory with the external application-specific geographical
features provided by the user. In the first case, the obtained important places are
still expressed in terms of geographic primitives and do not have any additional
information, so that the analysis is usually performed by domain experts using
visual analytics tools (Andrienko et al.
2007
). In the second step, the obtained
important places hold semantic information (id, location name) that can be used in
the data mining process (Alvares et al.
2007a
).
However, there are several problems with this approach: (1) An external
database of geographic features should be available. But even if it is available, the
algorithm can miss important places if the database is not complete. (2) The real
context of a stop is not known. For example a person may be waiting in a traffic
jam near a museum on the way to his/her work but the algorithm for finding
important places may identify the person as visiting the museum by extracting a
stay point (important place were a person spends considerable time) by inter-
secting the trajectory with the museum. (3) Since the data itself can have many
contexts at different time intervals, the important place found may not correspond
to the semantics that was attached to it (static semantics enrichment). (4) The
extracted semantics describe only the data they are attached to, and cannot be used
for other purposes.
Obviously, spatio-temporal data should contain more information to aid the
analyst in understanding the context of the data. Since photo-collection data
contains visual and textual information explaining the context of a photo, this data
has invaluable potential for the analysis of the geographical places to which photos
are geo-referenced, and the understanding of events that happen in the place where
the photo was taken.
In this paper, we provide a conceptual foundation for the analysis of events and
places using geotagged photo collections. We claim that a semantic enrichment of
the spatio-temporal data should use additional components available in the data
and take into account the temporal aspect. We define several types of semantic
