Discovering Collective Converging Groups of Large Scale Moving Objects in Road Networks

Abstract

Group pattern mining based on spatio-temporal trajectories have gained significant attentions due to the prevalence of location-acquisition devices and tracking technologies. Representative work includes convoy, swarm, travelling companion, gathering, and platoon. However, these works based on Euclidean space cannot handle group pattern discovery in non-planar space, such as urban road networks. In this paper, we propose a new group pattern, named converging, and its mining method in road networks. Unlike the aforementioned group patterns, a converging indicates that a group of moving objects converge from different directions for a certain time period. Motivated by this, we formalize the concept of a converging based on cluster containment relationship. Since the process of discovering convergings over large scale road network constrained trajectories is quite lengthy, we propose a density clustering algorithm based on road networks (DCRN) and a cluster containment join (CCJ) algorithm to improve the performance. Specifically, DCRN adopts the well-known filter-refinement-verification framework for efficiently identifying core points, which utilizes the upper bound property for \(\varepsilon \)-neighbourhood of point set on an edge to dramatically reduce the candidate core points. To process the neighbourhood queries efficiently, we further develop a vertex-neighbourhood based index, which precomputes the \(\varepsilon \)-neighbourhoods of all vertices, to facilitate neighbourhood queries of all points in road networks. In addition, to process the CCJ efficiently, we develop a signature tree based on road network partition index to organize the clusters in road networks hierarchically, which enable us to prune enormous unqualified candidates in an efficient way. Finally, extensive experiments with real and synthetic datasets show that our proposed methods achieve superior performance and good scalability.