| Summary: | Emerging trends in urban mobility have accelerated the need for effective traffic management and prediction systems. Simultaneously, the widespread adoption of GPS-enabled mobile devices has opened radical new possibilities for such systems. Motivated by this development, this thesis proposes an end-to-end streaming approach for traffic management that encompasses a novel prediction model. The stream processing is achieved by a sliding window model. In particular, the approach performs online 1) management of the current evolving trajectories, 2) incremental mining of closed frequent routes and 3) prediction of near-future locations of the moving objects based on the current object trajectories and historical frequent routes. The approach proposes storage of closed frequent routes and all possible turns a moving object can make at a junction, in a FP-tree like structure. This structure is created on the-fly from the buffered contents of each constituent window of the trajectories stream and then used to determine probabilistic future locations of each moving object. It additionaly calculates the densities of moving objects and parked objects for the entire road network. The prototype implements the approach as extensions to SCSQ - a data stream management system (DSMS) developed at UDBL. SCSQ is an extension of Amos II which is an extensible, mainmemory OO DBMS. The solution utilizes SCSQ’s stream manipulation and windowing capabilities coupled with Amos II’s functionality to efficiently store, index and query frequent routes for prediction. The approach is empirically evaluated on a large real-world data set of moving object trajectories, originating from a fleet of taxis, showing that detailed closed frequent routes can be efficiently discovered and used for prediction.
|
|---|
