Spatio-temporal fleet detection and Points of Interest identification for industrial/commercial vehicles

In recent years, technological advancements in many crucial fields have had an impact on the world of transport systems, giving birth to smart mobility. As more and more devices are connected to the Internet, also vehicles have been subject to transformations regarding connectivity: employing cutting-edge technologies, such as 4G/LTE/5G, vehicles are now able to provide real-time information to management services. Additionally, telematics companies working in industrial settings aim at providing services to thousands of operators, manufacturers and teams by processing the collected data. Finding fleets - groups of vehicles, owned or operated by a single entity - is therefore crucial to remotely monitor and coordinate the involved vehicles, optimize transportation services, enhance operational efficiency, and improve overall service quality. This master's thesis seeks to explore the pivotal role of fleet detection and enhancement of fleet management services in such context, with a particular emphasis on the collaborative efforts with Tierra S.p.A., a distinguished provider of telematics solutions. The primary objective of this thesis is to develop an algorithmic framework capable of detecting fleets of vehicles operating within the same geographical area, studying their temporal evolution, and classifying relevant Points of Interest (PoIs) such as rest areas and work zones. Leveraging geographical position data provided by vehicles over time, the proposed approach works in the research scope of graph analysis, managing data and constructing a graph representation of fleet interactions based on continuous geographical proximity. To detect fleets, two community detection algorithms – Louvain and Leiden – were studied. The analysis of the temporal evolution of fleets by means of an algorithm inspired by the MONIC framework allowed to provide valuable insights into fleet dynamics over time. Furthermore, the methodology extends beyond fleet detection to also identify PoIs by applying DBSCAN to the set of vehicle locations during interactions and to consequently classify them. Graph-based methodologies have emerged as effective techniques for fleet detection on industrial vehicles, complemented by clustering algorithms that efficiently identify fleets' PoIs.