Early detection of highway congestion from probe car data

Abstract

Highway congestion is an increasingly pressing societal problem, both in terms of cost (manyproductive hours lost) and safety (highway congestion increases the risk of accidents). While thereis a plethora of research on detecting and predicting traffic flow state from floating car data fromdata generators such as in-car navigation systems, little research has been done on how more de-tailed vehicle-generated data such as available on the vehicle CAN-bus (breaking, steering, etc.)could be translated into earlier or better quantification of the traffic flow state. The hypothesisunderlying my research is that the the data generated by participating cars can be modelled asa complex system of which the spatio-temporal complexity can be quantified; that a rise in thespatio-temporal complexity could be an early indicator of perturbed traffic flow; and that theaggregate patterns of multiple participating cars on a given road segment could in turn be mod-elled as such a system where the rise in spatio-temporal complexity is a good measure of thecongestion-proneness of the traffic condition. This information could be fused with historical dataon congestion probabilities to provide better congestion prediction with applications towards co-operative (cooperative adaptive speed control) or externally managed (variable message signs)traffic management strategies.