Scalable selective traffic congestion notification

Congestion is a major problem in most metropolitan areas. Systems that can in a timely manner inform drivers about relevant, current or predicted traffic congestion are paramount for effective traffic management. Without loss of generality, this paper proposes such a system that by adopting a grid-based discretization of space, can flexibly scale the computation cost and the geographic level of detail of traffic information that it provides. From the continuous stream of grid-based position and speed reports from vehicles, the system incrementally derives 1) statistics for detecting directional traffic congestions and 2) model parameters for a time-inhomogeneous, Markov jump process that is used to predict the likelihood that a given vehicle will encounter a detected directional congestion within the notification horizon. A simple but efficient SQL-based prototype implementation of the system that can naturally be ported to Big Data processing frameworks is also explained in detail. Empirical evaluations on millions of object trajectories show that 1) the proposed movement model captures the topology of the underlying road network space and the directional aspects of movement on it, 2) the congestion notification accuracy of the system is superior to a linear movement model based system, and 3) the prototype implementation of the system (i) scales linearly with its input load, notification horizon and spatio-temporal resolution and (ii) can in real-time process 1.14 million object trajectories.