Database Reference
In-Depth Information
selection of movements (in this case the trajectories with a specific average
speed
s
) the average length observed seems to capture well the behavior of the
observed trips, because the variance is low. In the Milano2007 data set, the plot
shows how the distance traveled grows linearly with speed, as expected, only
up to 80 km
/
h, while it decreases for higher speed. In the Pisa2010 data set,
the distance traveled grows linearly up to 110 km
/
h, with a low slope between
20 and 40 km
/
h. The plots show also the number of trips for each speed value:
the high diversity of lengths for speeds beyond 130 km
/
h (the highest speed
limit in Italy) is due to the low number of travels with that velocity and can be
considered as noise, coherently with the intuition that very fast trips take place
in particular situations of light traffic, typically at night.
We learned two lessons from our basic analytical explorations. First, all
statistics confirmed that there is a huge complexity represented in the data, a
wide variability of individual mobility behaviors that cannot be fully under-
stood in their diversity by looking only at macroscopic, global measures and
laws. Second, we realized that the basic spatio-temporal statistics are not well
suited to support the discovery and analysis of
movement patterns
, because
the very nature of a trajectory requires a deep understanding of the internal
dynamics of movement and their relations with the context. For these par-
ticular aspects, we exploit the mobility data mining methods introduced in
Chapter
6
.
10.4 Analysis of Movement Behavior
To answer the questions proposed in Section
10.2
, a complete mobility knowl-
edge discovery process centered around the trajectory concept is needed. Such
a process should be powered by a suitable system with the aim of supporting
interactive, iterative visual exploration of the analytical results, thus enabling the
analyst to combine different forms of knowledge and drive the analysis toward
the discovery of interesting movement patterns. An instance of a mobility knowl-
edge discovery process has been introduced in Chapter
7
. In this section we show
how the mobility data analysis tools are able to provide answers to the questions
discussed.
10.4.1 Origin-Destination Matrix Exploration
As stated in Section
10.1
, origin-destination (OD) matrix models provide a sim-
ple and compact representation of traffic dynamics, by abstracting detailed actual
movements by means of aggregation in flows between two regions. While the
traditional OD matrices are modeled by statistical analysis of surveys, sample
observations, and continuous refinements of the original models, the large avail-
ability of sensed tracks from real vehicles enables the automatic extraction of
