Index-supported pattern matching on symbolic trajectories

Abstract

Recording mobility data with GPS-enabled devices, e.g., smart phones or vehicles, has become a common issue for private persons, companies, and institutions. Consequently, the requirements for managing these enormous datasets have increased drastically, so trajectory management has become an active research field. In order to avoid querying raw trajectories, which is neither convenient nor efficient, a symbolic representation of the geometric data has been introduced. A comprehensive framework for describing and querying symbolic trajectories including an expressive pattern language as well as an efficient matching algorithm was presented lately. A symbolic trajectory, basically being a time-dependent symbolic value (e.g., a label), can contain names of traversed roads, a speed profile, transportation modes, behaviors of animals, or cells inside a cellular network. The quality and efficiency of transportation systems, targeted advertising, animal research, crime investigation, etc. may be improved by analyzing such data. The main contribution of this paper is an improvement of our previous approach, featuring algorithms and data structures optimizing the matching of symbolic trajectories for any kind of pattern with the help of two indexes. More specifically, a trie is applied for the symbolic values (i.e., labels or places), while the time intervals are stored in a one-dimensional R-tree. Hence, we avoid the linear scan of every trajectory, being necessary without index support. As a result, the computation cost for the pattern matching is nearly independent from the trajectory size. Our work details the concept and the implementation of the new approach, followed by an experimental evaluation.