PA-Tree: A Parametric Indexing Scheme for Spatio-temporal Trajectories

Abstract

Many new applications involving moving objects require the collection and querying of trajectory data, so efficient indexing methods are needed to support complex spatio-temporal queries on such data. Current work in this domain has used MBRs to approximate trajectories, which fail to capture some basic properties of trajectories, including smoothness and lack of internal area. This mismatch leads to poor pruning when such indices are used. In this work, we revisit the issue of using parametric space indexing for historical trajectory data. We approximate a sequence of movement functions with single continuous polynomial. Since trajectories tend to be smooth, our approximations work well and yield much finer approximation quality than MBRs. We present the PA-tree, a parametric index that uses this new approximation method. Experiments show that PA-tree construction costs are orders of magnitude lower than that of competing methods. Further, for spatio-temporal range queries, MBR-based methods require 20%–60% more I/O than PA-trees with clustered indicies, and 300%–400% more I/O than PA-trees with non-clustered indicies.