Database Reference
In-Depth Information
OD matrices. The suggested procedure is based on the traditional transportation
science approachwhere a spatial tessellation is used to generalize and summarize
movements. Starting from a given spatial tessellation, each GPS-tracked move-
ment is mapped to its corresponding origin and destination, that is, the former
is the region where the trip begins and the second is the region where the trip
stops. This kind of representation loses the focus on
how
people move, that is,
by which routes, and maintains only the information of the origin and the desti-
nation. Depending on the time interval considered, it is possible to reconstruct a
OD matrix for different time periods, allowing a precise characterization of the
evolution of traffic demand during time.
For the mobility data analyst, the OD matrix represents a valuable tool to
explore mobility data of a region, since it helps to reveal relevant flows and time
intervals. For example, to explore the main flows from the city center toward
the suburbs, we start by considering the administrative borders of Milan and its
adjacent municipalities (see Figure
10.5
a). A visual interface may enable the
analyst to disentangle the complexity of the model by exploring relevant flows
on the screen. There exist several methods to visualize and explore OD matrices
(see Chapter
8
for a review of visualization methods for flows); as an example,
Figure
10.5
b shows the visual interface provided by the M-Atlas system. In our
analysis, we focus on the flows leaving the city center of Milan toward the north
east suburbs.
10.4.2 Most Popular Itineraries from the City Center to Suburban Areas
Once we have selected a relevant set of flows, we can focus the analysis on the
individual trips associated to them.
The resulting trajectories are presented in Figure
10.6
a. Despite the fact that
all these trips originate in the city center and end in the northeast suburbs,
a broad diversity is still evident. In order to understand which are the most
popular itineraries followed by the selected travels, we apply an algorithm that
automatically detects significant groups of similar trips. In particular we use
the density-based clustering algorithm with the
Spatial Route
distance function
introduced in Chapter
6
. Given two trajectories, the route similarity function
returns a numeric estimation of their diversity: if the trajectories are equal it tends
to zero, otherwise it tends to infinity. A route is relevant for the mobility analyst
if it is followed by many vehicles. The clustering algorithm selects effectively
groups of trajectories with similar paths and thus provides a selection of frequent
routes. Trajectories that do not belong to any group are labeled as noise, and the
user might decide to discard them or, in some particular cases, to analyze them
separately from the others.
The clustering algorithm produces a set of clusters, each of which can be
visualized by means of a thematic rendering where the trajectories in the same
