Database Reference
In-Depth Information
(a)
(b)
(c)
Figure 11.6 Outlier detection.
are the grey boxes defined for some key positions of median trajectory linked
together. The limits of the spatio-temporal channel are outlined in light grey. Two
examples of outliers' trajectories (dark grey) getting out of the spatio-temporal
channel are presented in this figure. The first one presents a late temporal outlier
trajectory. The second one highlight a spatial outlier trajectory (right sided).
This spatio-temporal pattern must be computed for each HGT. As new positions
are frequently acquired by the system, this spatio-temporal channel could be
improved by updating it periodically.
Quality of the set of patterns depends on the precision of the ZOI graph
and the set of mobile object types. This quality could be verified if the spatial
and temporal distributions of positions of each
C
m
p
i
are unimodal. If several
modes appear, a new analysis can be carried out to split the set of mobile objects
according to types or to add new ZOI in the graph.
11.2.5 Outlier Detection
For each cluster, the associated spatio-temporal pattern splits the set of trajectory
positions in the outlier position group and the usual position group. For a new
vessel position, this knowledge could be useful to detect and to qualify this posi-
tion. Therefore, this section suggests that we combine the knowledge database
and the production database to obtain an inductive database and to detect the out-
lier positions in real time. Let's consider a new position
p
received. The position
qualification process is decomposed into three steps (illustrated in Figure
11.6
):
Trajectory extraction from the last ZOI encountered by the mobile object
to
p
,
