A framework for efficient multi-attribute movement data analysis

Abstract

In the first two decades of this century, the amount of movement and movement-related data has increased massively, predominantly due to the proliferation of positioning features in ubiquitous devices such as cellphones and automobiles. At the same time, there is a vast number of requirements for managing and analyzing these records for economic, administrative, and private purposes. Since the growth of data quantity outpaces the efficiency development of hardware components, it is necessary to explore innovative methods of extracting information from large sets of movement data. Hence, the management and analysis of such data, also called trajectories, have become a very active research field. In this context, the time-dependent geographic position is only one of arbitrarily many recorded attributes. For several applications processing trajectory (and related) data, it is helpful or even necessary to trace or generate additional time-dependent information, according to the purpose of the evaluation. For example, in the field of aircraft traffic analysis, besides the position of the monitored airplane, also its altitude, the remaining amount of fuel, the temperature, the name of the traversed country and many other parameters that change with time are relevant. Other application domains consider the names of streets, places of interest, or transportation modes which can be recorded during the movement of a person or another entity. In this paper, we present in detail a framework for analyzing large datasets having any number of time-dependent attributes of different types with the help of a pattern language based on regular expression structures. The corresponding matching algorithm uses a collection of different indexes and is divided into a filtering and an exact matching phase. Compared to the previous version of the framework, we have extended the flexibility and expressiveness of the language by changing its semantics. Due to storage adjustments concerning the applied index structures and further optimizations, the efficiency of the matching procedure has been significantly improved. In addition, the user is no longer required to have a deep knowledge of the temporal distribution of the available attributes of the dataset. The expressiveness and efficiency of the novel approach are demonstrated by querying real and synthetic datasets. Our approach has been fully implemented in a DBMS querying environment and is freely available open source software.