The growing popularity of location-enabled tracking devices has triggered a new sense of interest and enthusiasm for creating appropriate datasets or databases, and new approaches for carrying out data analytics. Demand and need for developing capabilities for tracking collective motion-related activities for objects--in isolation, as well as in groups or clusters--are on the rise. This paper showcases new approaches for analyzing trajectories of such temporally constrained clusters from different types of datasets.
One of the notable highlights of the work is the development of a new novel indexing structure, ReTraTree, designed to include an efficient indexing scheme for representing large dynamic moving object databases (MODs). The structure is designed with four levels of hierarchy. Two upper levels operate on the temporal dimension of data, while the third level engages the spatiotemporal dimension. The fourth level contains the actual archived data. The first three levels reside in primary memory and the fourth one in secondary memory. The description is rich with illustrations revealing the semantics of the architecture of the data structure.
The other major highlight is the introduction of a new QuT-Clustering algorithm that can be applied to the ReTraTree data structure. The discovery of clusters having the longest pattern is made feasible through a combination of merge and append operations applied to the query results. The authors claim this to be a simple (and unique) approach for identification of clusters. Experiments were carried out over both synthetic MOD (SMOD) and real datasets (IMIS and GeoLife). Experimental evaluation was done by creating a bundle of queries over randomly selected time windows and executing them in random sequence. The performance of the new approach scaled better over rival approaches.
By providing an outline of the algorithm, and complexity analysis of the entire framework that includes the new algorithm, the work is sure to catch the interest of researchers working on similar temporally constrained datasets.