Information Technology Reference
In-Depth Information
in this topic and further related work in the area, GIScience has made significant
progress in seeking structure in movement data. Algorithms have been proposed to
cluster similar trajectories and segment trajectories, to compute home ranges and
to find patterns. However, attaching such structures to colloquial names (“flock”,
“herd”, “convoy”, “single file”), does not necessarily mean that a found pattern
corresponds to an actual herd of cows or trucks indeed moving in a convoy. In short,
whereas finding structure is easy, the following semantic annotation and enrichment
of the found structures is and remains much harder.
Progress in this second but crucial CMA step is slow for two main reasons. First,
most work so far was tools-driven and methods-driven, but not problem-driven.
Second, most movement data used so far consisted of bare trajectories without any
form of semantic metadata. For instance, when cows were tracked, no information
was captured about the social structure of animals. Or, when observing two users of
a mobile phone in the same cell for half an hour, we have in general no information
knowing for certain that they actually met for coffee. That means, even when aiming
at a semantic evaluation of proposed methods, finding appropriate data to do so
is difficult. Nevertheless, when aiming at really understanding the processes and
events controlling the observed movement, then bridging the semantic gap between
formal representations of patterns and structures and their actual meaning grounded
in contextual expert knowledge is key. Initial work combining movement data with
social media data (for example, applications allowing users to “check-in” at points
of interest) offer a promising route for enriching raw trajectories with user-generated
semantics (Sui and Goodchild
2011
).
Some research fields concerned with CMA have proposed promising work
addressing this gap. Regarding data capture, recent work in movement ecology no
longer just monitors the location of observed animals, but uses multi-sensor devices
simultaneously tracking physical properties of the individuals (for example, speed,
acceleration, physiological variables). The underlying rational is that activities can't
be inferred from location fixes alone, irrespective of how densely they are sampled.
A second development in CMA specifically aims at understanding movement by
understanding its embedding in its enabling and constraining geospatial context (
P3
.
Laube et al.
2007
). Cows that are co-located for some time may not belong to the
same herd at all but just be kept in a fenced area. Similarly, understanding com-
muter patterns without studying the transportation infrastructure obviously makes
little sense. Again, given the strong influences of geometry and topology oriented
researchers, it may be little surprising that so far most CMA focuses on shape and
arrangement of trajectories and less on the development of context-aware movement
analysis techniques.
