Trajectories and Their Representations - Mobility Data - page 14

in Figures 1.1 and 1.5 : The trajectories (1) have first been segmented into Stop
and Move episodes, (2) their Move episodes have been annotated with a new
annotation, the main transportation means for this move segment, and (3) their
Stop episodes have been annotated with two new annotations: the geo-object
(POI) where the stop took place and (POI) the activity of the tourist during this
stop. Figure 1.5 provides also an alternative annotation of the Stop episodes:
the types of the POI associated to the stop (hotel, museum, restaurant, etc.)
instead of the POI itself (the hotel Zola, the museum Le Louvre, the restaurant
Babylone, etc.).
Once the best-fitting representations of the trajectories have been built, appli-
cation analysts can use them to extract all kinds of statistical and higher-level
knowledge useful to the application.
1.4 Trajectory Patterns and Behaviors
Section 1.3 discussed how to enrich the raw trajectories with the related contex-
tual data to come up with semantically rich trajectories. This section discusses
the concepts involved in the process that extracts relevant semantic knowl-
edge from the trajectories. Since long ago, researchers have developed novel
techniques to extract knowledge taking into account the spatio-temporal speci-
ficity of movement data. These techniques support learning from trajectories far
beyond retrieving factual data about specific moving objects (e.g., where was the
car 345FT92 at time t ?) and computing statistics about populations of moving
objects (e.g., how many cars per hour travel this road on weekdays?).
Of vital importance for a large number of applications is the identification
of the significant trends shown by a population of moving objects. Sociological
studies, for example, may aim at comparing commuters' shopping habits versus
shopping habits of noncommuters. Trajectory analysis reveals which persons
qualify as commuters and identifies their favorite shopping places. Similarly,
analysis of tourists' trajectories may detect trends in tourist behavior that provide
important information to tourist agencies to optimize their offers.
A significant trend can be identified as a set of trajectory characteristics that
repeatedly appear in the set of trajectories under consideration. Most frequently,
trends are “found” using a knowledge extraction tool, usually applying data
mining techniques. The datamining community uses the term“pattern” to denote
the findings from the extraction, and “frequent pattern” to denote those patterns
that appear frequently enough in the source data to be considered potentially
interesting for the application at hand. For example, “The trajectory ends at the
same place it began” is a trajectory characteristic that can be denoted as a Loop
pattern. The pattern identifies trajectories whose spatial trace, as a whole, forms
a loop. Its definition relies on the spatio-temporal positions Begin and End, and
nothing else. We call it a spatio-temporal pattern .