Prediction of Red Light Running Based on Statistics of Discrete Point Sensors

A probabilistic model is proposed to predict red light running (RLR) for collision avoidance systems at arterial signalized intersections. This RLR predictor consists of an arrival time estimator and a statistical predictor of vehicle stop-and-go maneuvers with two discrete point sensors (capable of measuring speed). In addition, unlike most prediction models, which are designed to minimize mean errors, this model identifies two types of error: the false alarm and the missed report. The capability of distinguishing these two types of error is crucial to the effectiveness of RLR-related collision avoidance systems. Therefore, the Neyman–Pearson criterion is employed: it keeps the false-alarm rate lower than a given threshold while at the same time minimizing the probability of missing error. To quantify the trade-off between these two types of error in the system design, a system operating characteristics (SOC) function is defined. The system parameters are determined by using an offline supervised parameter-setting procedure in which training data are collected from a field intersection in the San Francisco Bay Area in California with Autoscope video cameras. Effectiveness of the proposed model and its prediction algorithm are demonstrated by the collected field data. The theoretical system performance predicted by the SOC curve is matched with the evaluated performance by means of data collected from field intersections. For example, at a preset false-alarm rate of 3%, the correct prediction rate of RLR for three approaches of the field intersection ranges from 63% to 80%.