Community-based vehicular networks for traffic information systems

Traffic information systems (TIS) monitor road traffic conditions with the goal of improving the efficiency of road networks. Given the ever increasing number of cars, TIS are becoming more critical, as they help reduce traffic congestion which have large negative social, economical, and environmental impact. Current TIS have several drawbacks. First, there are insufficient sources of traffic data, that often require either expensive sensing infrastructure or a contract with a network provider owning GPS data. Secondly, they are often limited by a centralised architecture introducing both delays in delivery of real-time data and a single-point of failure. Finally, even after collecting traffic information, they provide no easy way to effectively relay information to drivers or apply traffic management strategies. New vehicle-to-vehicle (V2V) communication technologies can potentially aid in overcoming these limitations. Using V2V, vehicles can communicate with each other and exchange traffic information within an ad hoc vehicular network (VANET). VANETs could drastically change the nature of traffic systems, moving from a centrally-controlled to a self-organised complex system covering every road segment.This thesis studies the potential of VANETs in the context of traffic data collection and traffic management using a multidisciplinary approach. We combine knowledge of networking protocols, distributed systems, and traffic theory to propose novel solutions for TIS. First, we propose a TrafficEQ system---a VANET-based TIS, that uses pure V2V communication to collect and disseminate traffic information among vehicles. Moreover, TrafficEQ provides intelligent route guidance based on a probabilistic route choice strategy. The system is evaluated using a simulation platform developed with realistic real-world scenarios. We demonstrate that TrafficEQ deals with challenging non-predictable traffic congestion better than traditional route guidance systems can. Secondly, we propose grouping vehicles into communities by similarity of their mobility patterns to improve traffic congestion detection and analysis. To facilitate this, we present the Crowdz algorithm---a community detection algorithm for VANETs. Extensive analysis on large-scale vehicular networks show that detected communities are more stable than those in a competitive algorithm. Lastly, we introduce an application which uses communities to detect and analyse traffic congestion. Simulation experiments show that identification of congested communities can help in applying appropriate control strategies