Database Reference
In-Depth Information
Figure 11.4 Clusters of positions and spatial pattern.
Once every position is matched, the coordinate and the timestamp of P m i are
updated, by computation of median X ( X ), median Y ( Y ), and median timestamp
( t ). Amedoid approach is also possible but requires more time for similar results.
Assignment and update steps are repeated until the distance (Frechet distance or
average distance) between two consecutive points reaches a minimal threshold
value.
As the studied mobile objects move in an open area, some of them can move
away from the main trajectory. Normal temporal or spatial deviations must
be distinguished from outliers. Two channels are computed to distinguish the
spatio-temporal outliers. First, the spatial channel is defined. Once the median
trajectory is computed, a statistical density analysis can be performed on every
cluster of matched positions (
C m p i ). These clusters are split into two subsets
of positions,
R p i (right sided), according to their side to
the median position P m i using the P m i heading. Then, spatial distances between
positions from
L p i (left sided) and
L p i and the P m i are computed. After a statistical analysis, the
Search WWH ::




Custom Search