Information Technology Reference
In-Depth Information
The proposed algorithm shows the spatio-temporal algorithm presented by this
paper. This algorithm basically carries out the first part of TRACLUS with no change
at all, which is the partitioning phase. It then begins clustering the segments into
groups but instead of using the weighted distance function of TRACLUS, which uses
the parallel, angular and perpendicular distances, the GenSTLIP function is
implemented to give the algorithm the spatio-temporal sense. This function helps
compare line segments for both their spatial and temporal dimensions.
GenSTLIP is used a little bit differently than it was introduced. GenSTLIP
originally works on whole trajectories, giving a final value indicating how similar two
trajectories are. The same concept is used by this algorithm, but instead of comparing
trajectories as a whole, it does the same calculations but on each two pair of segments,
since TRACLUS already does the partitioning part. It is believed that in this way it is
possible to get the advantage of both; the advantage of partitioning as TRACLUS
suggests discovering similarities between sub-trajectories which is better than
comparing trajectories as a whole, and the advantage of the spatio-temporal function
of GenSTLIP.
4
Experimental Evaluation
The proposed spatio-temporal algorithm is conducted on three different data sets. The
data sets used are those the same used to test TRACLUS but added to them the time
dimension. We purposely use those data sets to be able to see the difference when
those data sets are analyzed spatio-temporally instead of just spatially.
4.1
Hurricane Data Set
The Hurricane data set [9] consists of a set of Hurricanes in the Atlantic Tropical
region of North America. This data set is known as the Best Track. This data sets
contains Atlantic hurricanes from years 1950 through 2006. It consists of 608 un-
partitioned trajectories. Each trajectory is made up of several points representing it.
Each point consists of three parameters (x-coordinate, y-coordinate, time).
4.2
Animal Movement Data Sets
Two different animal movement data sets have been used in this paper to conduct the
proposed algorithm. The first data set represents elk movement in year 1993 while the
second data set represents deer movement in year 1995. Both these data sets have been
generated by the Starkey Project [10]. The elk data sets contains 33 un-partitioned
trajectories while the deer data set contains 32 un-partitioned trajectories.
All experiments are conducted on an i5-4200M 2.5 GHz CPU, with 4.0 GB of
main memory running on Windows 8.1. The algorithm is implemented in C++ using
Microsoft Visual Studio 2010. For graphical representation of each cluster, Matlab
R2010a is used.
Search WWH ::




Custom Search