Database Reference
In-Depth Information
[113]. In such a case, the shapes of the trajectories of the dif-
ferent objects may sometimes be quite different. The approach
discussed in [113] first uses an off-the-shelf spatial clustering al-
gorithm to partition the objects into clusters at each time-stamp.
This transforms the spatial trajectories into data which represents
membership of objects in clusters. Subsequently, a frequent pat-
tern mining-like approach is applied to this transformed data into
order to determine those objects which belong to the same spatial
cluster for a significant number of time-stamps. An Apriori-like
approach is used for this purpose, in combination with a number
of additional pruning tricks, which use the temporal characteristics
of the data. Since the consecutiveness of the membership informa-
tion is not used in the pattern-mining phase, the swarms are based
on significant levels of co-location at any period in time.
The problem of clustering is particularly useful from the perspective of
trajectory mining, because it provides summary information which can
be used for other applications. For example, the
TraClass
method pro-
posed in [105] uses two kinds of clustering in order to provide additional
summary information, which enables more effective classification. One
kind uses the characteristics of different regions in the clustering, but
it does not use the movement patterns. The other kind uses the char-
acteristics of different trajectories in the clustering. The two kinds of
clusters provide useful complementary information in the classification
process. It has been shown in [105], how this additional information can
be leveraged for a more effective classification process.
While clustering determines the
typical
movement patterns, a related
problem is that of determining unusual (or
atypical
) movement patterns
[105, 109, 110]. Such movement patterns are also referred to as outliers.
Another variation on the problem is the determination of periodic pat-
terns [114], which we wish to determine common patterns of movement
which repeat periodically in the trajectory data, or hot routes in road
networks [111]. A comprehensive range of trajectory mining techniques
have been developed in the context of the
MoveMine
project at UIUC
[115].
While much of this work has been performed in the context of ani-
mals, similar techniques can be generalized to the case of humans. Hu-
man movements are of course somewhat more complex, because of the
greater complexity of social interactions as compared to animals. Some
recent work has been performed on studying the trajectory patterns of
humans, which were collected from mobile phones [73]. It was shown
that human trajectories show a high degree of temporal and spatial
regularity, and each individual shows a highly time-independent charac-
