Prediction Method for the Occurrence Probability of Freeway Traffic Collisions

This paper presents a simple method for short term prediction of freeway traffic collisions based on the traffic flow theory. A model is proposed to estimate the probability of traffic collision occurrence, which is achieved by analyzing the real time traffic flow data from two neighboring traffic detectors. The proposed model indicates that the probability of traffic collision occurrence is positively related to two terms: a) the ratio of the upstream traffic density to the steady traffic density, and b) the speed difference between the upstream traffic and the downstream traffic. The empirical data from Dutch motorways is used to calibrate the model. The proposed model only needs fundamental traffic data to estimate traffic collision occurrence probability. It appears promising for the application in the on-line traffic incidents predictions in the future.