Indexing Large Trajectory Data Sets With SETIV
Prasad Chakka, Adam Everspaugh, Jignesh M. Patel
With the rapid increase in the use of
inexpensive location-aware sensors in a variety of new applications, large
amounts of time-sequenced location data will soon be accumulated.
Efficient indexing techniques for managing these large volumes of
trajectory data sets are urgently needed. The key requirements for a good
trajectory indexing technique is that it must support both searches and
inserts efficiently.
This paper proposes a new indexing mechanism called SETI, a Scalable and
Efficient Trajectory Index, that meets these requirements. SETI uses a
simple two-level index structure to decouple the indexing of the spatial
and the temporal dimensions. This decoupling makes both searches and
inserts very efficient. Based on an actual implementation in SHORE, we
demonstrate that SETI clearly outperforms two previously proposed
trajectory indexing mechanisms, namely the 3D R-tree and the TB-tree.
Unlike previously proposed trajectory indexing structures, SETI is a
logical indexing structure that uses existing spatial indexing structures,
such as R-trees, without any modifications. Consequently, database systems
that currently support R-trees can easily implement SETI, making it a both
a practical and an efficient choice for indexing trajectory data sets in
many existing database systems.