Trajectory Data Warehouses - Mobility Data - page 76

In the above query, function atMLine projects the time-dependent field to the
movement track of the episode, resulting in a time-dependent real. Then, function
mmax obtains the maximum temperature value during the episode.
The class of spatio-temporal OLAP and continuous field ( STOLAP-CF )
queries is the class that contains the queries expressed by Q agg augmented
with spatial types, time-dependent types, and field types. It follows that a con-
tinuous field data warehouse is a data warehouse that supports STOLAP-CF
queries.
4.6 An Example Trajectory DW: GeoPKDD
We showed in previous sections that individual trajectories can be represented
in facts and/or dimensions and that they can be aggregated and analyzed. An
alternative way of analyzing trajectory data, as we commented in Section 4.3 ,
consists in partitioning the space into regions (or road segments) and precom-
puting aggregated trajectory data relative to each partition. For example, we can
partition the space into regular squares and for each square compute the number
of trajectories at a given instant. This precomputation allows us to get rid of the
trajectories, and analyze them using traditional DWs. One relevant example of
this approach is the TDW developed in the GeoPKDD project. 1
The GeoPKDD TDW allows analyzing trajectory data without actually stor-
ing the trajectories themselves, but instead storing preaggregated measures
resulting from a complex ETL process that feeds the TDW. During this ETL
process, the sampled positions received by GPS-enabled devices are converted
into trajectory data and stored in a moving object database, using the trajectory
reconstruction techniques explained in Chapter 2 . The moving object database
also contains user profiles, spatial partitions, and temporal intervals.
After the reconstruction step, the TDW is fed with aggregate trajectory data
using either a cell-oriented or a trajectory-oriented ETL approach. The cell-
oriented approach searches for the trajectory portions that lie within the spatio-
temporal cells. Then, those portions are decomposed with respect to the user
profiles they belong to. On the other hand, the trajectory-oriented approach
looks for the spatio-temporal cells where each trajectory resides. Then, portions
of the trajectory that fit into each of those cells are computed, taking into account
the user profiles.
In such a TDW, the dimensions are typically organized as follows. The
temporal dimension is designed to range over equally sized time intervals, which
can be aggregated according to larger intervals as we move up in the dimension
hierarchy. The spatial dimension represents a partition of the space that defines
the cells (or the road segments) where measures are recorded. Further, the fact
1 http://www.geopkdd.eu