Back to Top

■ Corridor Learning Using Individual Trajectories

Nikolaos Zygouras, Dimitrios Gunopulos: Corridor Learning Using Individual Trajectories. In Mobile Data Management (MDM), 2018 19th IEEE International Conference on. IEEE. (to appear)
In this work, we proposed a pipelined approach for detecting a set of frequently accessed corridors from a vast collection of trajectories. Initially we applied a well known topic modelling technique to detect frequent sets of locations and then we derived the frequent corridors from the trajectories that accessed these locations.


The rapid development and commercialization of location acquisition technologies generates large trajectory datasets, that trace moving objects' trips. It is particularly difficult to interpret, analyse and monitor these datasets. In this work, we propose a new trajectory mining algorithm, for discovering paths that are frequently followed by the given trajectories, named as corridors. We claim that the moving objects, although they have different origins and destinations, they contain parts-\emph{subtrajectories} that follow common paths-\emph{corridors}. Detecting corridors from a collection of trajectories is extremely challenging due to the nature of the data (low sampling rates, different speeds, noisy measurements etc.). In this work we propose a pipelined algorithm that abstracts from the trajectories the underlying frequent paths, that represent the main mobility flows. Our approach first divides the space into smaller subareas, by detecting frequent sets of locations that co-occur in recurring trajectories. Then it learns a set of corridors at these areas. Finally, a set of paths is selected using a greedy approach that complies with the Minimum Description Length (MDL) principle.