Use of vehicle positioning data for arterial incident detection

This paper investigates the effectiveness of individual vehicle movement measures in the detection of incidents on urban arterial road segments. Several measures of varying measurement complexity, including average speed, running time and speed, and coefficient of variation of speed, are calculated from vehicle positioning data recorded at 1-s time intervals. The sensitivity of measures to differences in traffic condition is investigated, and the most sensitive measures are used as predictor variables in discriminant classification models. The results of these analyses are presented and discussed, including a simulation of the implementation of the classification models.