Trajectories and Their Representations - Mobility Data - page 12

Figure 1.5 A Stop/Move segmented representation of the tourist's trajectory of Figure 1.1 .
to the temporal data of the raw trajectories. For instance, June 17, 2012, was
Ascension Day, a public holiday in France.
All this information is conceptually associated with the spatio-temporal posi-
tions of the trajectory, but it would be space and time consuming to actually store
it for each position of the trajectory. Indeed, usually one does not characterize a
given position but a sequence of positions. For instance, when a tourist visits Le
Louvre museum, he or she may stay there for several hours, which may mean
thousands of consecutive recorded positions with the same annotation value:
“Le Louvre.” Therefore a common method consists in segmenting the trajectory
into maximum subsegments of spatio-temporal positions that are all associated
with the same value of a given expression whose range of values is a finite set
of annotation values. Each change of value of the expression signals the starting
of a new segment. The segments are called episodes and, instead of storing the
information with the position, it is stored with the episode. A common kind of
segmentation is the segmentation into episodes of kind Stop (segments of the
trajectory where the object roughly does not move) and Move (segments of the
trajectory where the object moves). It is a generic segmentation that relies only
on computation of the raw data. It is often based on the instantaneous speed of
the object, but the exact expression depends upon the application. For the tourists
scenario, the expression could be the Boolean expression: speed ≤ 1km/h.This
expression defines a Stop episode when the expression value is True, and a
move episode when it is False. Chapter 2 provides more details on trajectory
segmentation methods.
As shown in Figure 1.5 for the tourist's trajectory of Figure 1.1 , segmenta-
tion produces a semantic representation that is more abstract than the contin-
uous representation it comes from. The continuous representation contains the
sequence of spatio-temporal positions of the trajectory, while a segmented rep-
resentation provides a semantic view of the trajectory as a sequence of episodes,
each one described by a tuple (time interval, annotation value). A segmented