Maritime Monitoring - Mobility Data - page 232

Table 11.3 Results for Filtering Process with 10 m Tolerance.

Trajectory % of Position % of Length

Vessel Duration Kept Kept (km)

Bindy 28 m 01s 14.0% 99.91%

(32/229) (11.284/11.294)

Port pilot boat 1 h 07 m 36 s 21.7% 99.82%

(122/562) (24.846/24.892)

AB Valencia 7 h 04 m 20 s 12.0% 99.98%

(279/2316) (175.07/175.109)

Tabl e 11.3 summarizes the filtering result. One can note that their lengths
are very close. This leads to a filtering process where more than 80% of the
received positions can be filtered. The performance of the filtering process is
likely to increase for large ships and decrease for small ships due to the intrinsic
characteristics of their navigation.

11.2.3 Trajectory Clustering Process

Once the trajectory concept is defined, different trajectory clustering techniques
can be used to determine homogeneous groups of trajectories. Some of them
are presented in Chapter 6 . Another technique based on the zone graph and
itineraries can be used to extract clusters from trajectories following the same
itinerary I . This set is called a homogeneous group of trajectories (HGT).
The first selection criterion of this approach is based on static information
such as the type of mobile objects; this information is provided by AIS messages
(Table 11.2 ). The second selection criterion is a geographical one. The first
position of the trajectory ( p b ) must be the only one within the departure zone
( Z D ) of the itinerary, and the last position of the trajectory ( p e ) must be the
only one within the arrival zone ( Z A ) of the itinerary. Taking into account the
frequency of trajectory samples and the speed of the mobile object, trajectories
that cross a zone of the graph should have at least one position within this zone.
The last selection criterion used is time. Some moving objects can follow this
itinerary periodically. These different trajectories can be distinguished using a
time interval. Finally, the trajectory should not intersect any other zone of the
graph G Z that does not belong to the itinerary I . All valid trajectories previously
extracted from the STDB compose the HGT to be analyzed.
Figure 11.3 c illustrates the extraction of the HGT of 500 passenger ships'
trajectories following the itinerary between Brest and Naval Academy (arc A-
Fof G Z ). Some density differences can be noticed on this HGT. This HGT