Querying and Mining Trajectory Databases Using Places of Interest

Abstract

The study of moving objects has been capturing the attention of Geographic Information System (GIS) researchers. Moving objects, carry- ing location-aware devices, produce trajectory data in the form of a sample of (Oid, t, x, y)-tuples, that contain object identifier and time-space informa- tion. Recently, the notion of stops and moves was introduced. Intuitively, if a moving object spends a sufficient amount of time in a certain geographic place (which we denote a place of interest of an application), this place is considered a stop of the object’s trajectory. In-between stops, a trajectory has moves. In this paper we study how moving object data analysis can ben- efit from replacing raw trajectory data by a sequence of stops and moves. We first propose a formal model and query language (denoted Lmo) to ex- press complex queries involving spatial data stored in a GIS, non-spatial data (stored in a data warehouse) and moving object data. This query language also supports different forms of aggregation. We then study the compression of trajectory data produced by moving objects, using the concepts of stops and moves. We show that stops and moves are expressible in Lmo and that there exists a fragment of this language (that can be expressed by means of regular expressions) allowing to talk about temporally ordered sequences of stops and moves. We use this fragment to perform data mining over tra- jectory data. We present an implementation and a case study, and discuss different applications of our approach.