Querying trajectories

Retrieval tasks

Description

The Data Management group has targeted points-based search where trajectories are returned based on their proximity to a set of query points; i.e., given a collection of trajectories and a set query points, the goal is to retrieve the top-k trajectories that pass as close as possible to all query points. We advanced the nearest-neighbor based state-of-the-art by proposing a novel and more efficient, spatial range-based approach. In addition, a practical variant of this points-based search was proposed, which additionally takes into account other qualitative characteristics of the searched trajectories, e.g., the temporal span. Besides one-time or snapshot search. our recent work also introduced continuous points-based trajectory search where the query is long-standing and the result set must be maintained whenever updates to the query parameters and/or the data, i.e., the trajectories, occur.

Publications

 

Path finding

Description

This line of work is captured by Prof. Bouros' doctorate studies. Prof. Bouros addressed new challenges that arise in path-finding problems, given the availability of trajectory collections; in prarticulal, whether path queries traditionally targeting graphs can be posed on trajectory collections, and even more importantly, if the evaluation of these queries can be enhanced by the special characteristics of the trajectories. For instance, a trajectory can be seen as a set of precomputed answers. Under this perspective, a novel framework was proposed for evaluating path queries on large disk resident collections that are frequently updated by adding and removing trajectories. The thesis introduced two evaluation paradigms that enjoy the benefits of search algorithms (i.e., fast index maintenance) while utilizing transitivity information to terminate the search sooner. In addition, efficient indexing schemes and appropriate updating procedures were also introduced.

Publications