Database Reference
In-Depth Information
50
45
40
35
30
25
20
15
1
5
0
0
5
10
Time (hours)
15
20
25
(a)
60
50
40
30
20
10
0
0
5
10
Time (hours)
15
20
25
(b)
(c)
Figure 10.12 Temporal distribution of (a) arrivals to and (b) departures from the arena
area: arrivals peak from 5 P.M. through 8 P.M., and departures peak from 10 P.M. through
midnight. Arrivals are spread over several hours, while departures occur soon after the end
of the match. (c) clusters of trips leaving the arena after the football match. Clusters are
highlighted by shades. The largest cluster performs short range trips or take the road ring,
either toward the northeast or southwest.
Going deeper in the analysis, we might want to understand when and how
attendees reached and left the event location. First, the arrival and departure
time of the each car
v
parked in the arena area during the day is approximated
considering, respectively, the ending point of the incoming trajectory and the
starting point of the outgoing trajectory of
v
. The distribution of arrivals and
departures during the day is depicted in Figure
10.12
a,b. We further analyze the
return travels of the attendees after the match, in order to detect the main escape
routes - notice that they might differ from the routes planned (for example) by
public authorities, either in shape, frequency, or timing. We apply clustering to
the trajectories leaving the arena area between 10 P.M. and midnight, obtaining
the clusters shown in Figure
10.12
. The detected escape routes are relevant
information for a mobility manager to enact countermeasures to prevent possible
congestion.
10.4.5 Mobility Prediction
The prediction of traffic congestions represents a challenging task for urban
mobility managers. The following experiments are aimed at showing how to
predict future areas of dense traffic that may lead to traffic congestions. For
this task we use the
WhereNext
location prediction algorithm (introduced in
Chapter
6
) and run the experiment on the Pisa2010 data set, which covers a
larger area and a longer temporal interval compared with the Milano2007 data
set. This is particularly useful in prediction tasks because the training and test
phases use a richer data set. Here, we selected a subset of the entire Pisa2010
