Database Reference
In-Depth Information
Matching process between this trajectory and the median trajectories of a
pattern, and
Spatio-temporal comparison between
p
and selected pattern.
In the first step, the database is queried to select the start position from the
trajectory. This position is the last one of the mobile object inside the surface of
one ZOI. Positions between
p
and this departure position are timestamp ordered
to define a trajectory path. This last one does not link two ZOIs, consequently,
it is called a partial trajectory (
T
p
). In Figure
11.6
, the last ZOI is
A
and the
start position is (
b
). The partial trajectory is the dashed polyline. The second
step must match
T
p
with part of a median trajectory. This matching can be done
according to:
The type of the moving object,
The geometry of the partial trajectory,
The set of median trajectories from the departure ZOI, and
Information about the course of the moving object to destination.
Unfortunately, information about the destination is often false or unknown,
so only the type of vessel and geometry properties can be used. In order to match
two linear geometries, the Frechet discrete distance is selected as it allows partial
matching processes. Frechet distance gives the maximal distance between two
lines. The Frechet discrete distance applied to two discrete trajectories (ordered
set of points) can match trajectories together, preserving order of their points. Alt
and Godau (1995) demonstrate the advantage of this measure. Devogele in 2002
proposed to enhance this distance in order to compute the distance between a
line and a homologous part of another line. This partial discrete Frechet distance
(
dPdF
) is very useful to match a trajectory where only the departure is known.
Thanks to this
dPdF
, the distance between
T
p
and median trajectories from the
same departure ZOI can be computed. Only the spatio-temporal patterns for the
same type of this object are taken into account.
T
p
can be partially matched
with one median trajectory (
T
) where
dPdF
(
T
p
;
T
) is lower than the
dPdF
with
other median trajectories plus a threshold.
dPdF
(
T
p
;
T
) must also be less than
a maximal value. In the example, the distance between
T
p
and two median
trajectories (from ZOI
A
to
E
and from
A
to
F
) are computed. The second
distance is the lowest, so
T
p
is matched with median trajectories from
A
to
F
.
Finally, the position
p
could be qualified according to the selected pattern.
The relative time of
p
from departure ZOI is employed to infer the spatio-
temporal channel from the knowledge database. The 3D channel is cut at this
timestamp and the space is split into five areas (right, left, usual, late, and early).
Qualification of
p
is given by the area that contains
p
. For example, the spatial
channel of the matched pattern is limited with dark grey lines and the usual area
at the relative time of
p
is the grey area. Position
p
is an outlier and is located in
