Continuous k-Similarity Trajectories Search over Data Stream

Abstract

Continuous k-similarity trajectories search (short for CKST) over data stream is a fundamental problem in the domain of spatio-temporal database. Let $\mathcal {T}$ be a set of trajectories, $T_q$ be the query trajectory. $T_q$ monitors elements in $\mathcal {T}$, retrieves k trajectories that are the most similar to $T_q$ whenever trajectories in $\mathcal {T}$ are updated. Some existing works study k-similarity trajectories search over historical trajectory data. Few efforts could support continuous k-similarity trajectories search over data stream, but they cannot accurately measure similarity among trajectories.

In this paper, we propose a novel framework named SLBP (Score Lower-Bound-based Prediction) to support CKST. It is based on the following observation, that is, given $T_q$ and one trajectory $T_i\in \mathcal {T}$, if the distance between their corresponding last generated GPS point is large, the distance between $T_q$ and $T_i$ also may be large, and it cannot become a query result trajectory for a long time. In this way, we can predict the earliest time $T_i$ could become a query result. We develop a group of algorithms to support CKST via using the above property.