Near-Miss Collision Risk in Congested Traffic Flow

We present results of an investigation of the near-miss collision risk in traffic flow, that elevates observables, such as traffic density and velocity fluctuations, to a metric of accident precursor. Through rigorous use of concepts in statistical physics, we define the near-miss collision risk by means of the velocity state transition matrix, and propose a method of determining it from a dual definition of the velocity autocovariance function. We corroborate our method using both simulation-generated and crowdsourced vehicle velocity data. First, from exploring the phase-space of the Nagel-Schreckenberg cellular automaton simulation model, we find that the near-miss collision risk only occurs in congested flow, and is the greater the smaller fluctuations. Second, amidst the incessant rise of vehicle-caused fatality rates worldwide, the scalability of our approach to massively crowdsourced velocity data opens new venues for assessment and mitigation of actual vehicle collision.