Analysis and fast querying of mobility data

Mobility data represents a widely used term to indicate sources of information, possibly structured in many different ways - depending on technologies and formats used - describing the localization or the movement, in time and space, of sets of entities. Regardless of specific ways through which mobility data is represented, this class of information is nowadays pervasive since it is massively produced, processed and analyzed for many different purposes. The importance of mobility data is going to increase even further in the future, given the ever growing diffusion of the internet of devices and the progressive introduction of the internet of things paradigm. In this thesis we contribute to mobility data research by addressing separately two distinct problems. The first one is related to the on-line processing of streams of mobility data coming from massive amounts of moving objects, where such streams contain location updates and some kind of queries continuously and periodically issued by the objects. This problem is frequently met, nowadays, in the context of Location-Based Services (LBS) or Location-Based Social Networking applications (LBSN), even if one has to observe that the nature of the problem allows it to be possibly found in quite diverse domains, such as massively multiplayer online games, anti-collision detection systems, behavioural simulations and so on. More precisely, we focus on the problem of computing massive range and k-nearest neighbour queries, which represents the time-dominant phase of the whole processing. In order to tackle effectively the problem we exploit the remarkable - yet cheap - computational power of modern GPUs by introducing novel algorithms and data structures, and we prove the effectiveness of our solutions through an extensive series of experiments. The second problem relates to the domain of mobility data mining. In this context the main goal is to devise novel, off-line analyses able to extract previously unknown and interesting patterns from raw mobility data. This kind of research is, in general, very interesting since it allows to gain new insights on mobility data. We address the problem of detecting avoidance behaviours between moving objects from historical movement traces. To this end, we first introduce a framework which formally defines what is an avoidance behaviour between moving objects; subsequently, on the basis of such framework we provide an algorithm which is able to extract these patterns. Finally, we experimentally prove the effectiveness of our solution with real-world datasets.