The impact of motion prediction methods on surrogate safety analysis: A case study of left-turn and opposite-direction interactions at a signalized intersection in Montreal

Abstract

ABSTRACTSurrogate safety analysis is a proactive approach that relies on the observation of traffic events without a collision. To measure the proximity to collision, several safety indicators have been proposed in the literature. Most indicators like time to collision (TTC) rely on the choice of a motion prediction method such as the rarely justified prediction at constant velocity. Moreover, most current interpretations of continuous safety indicators use only a single value to qualify the whole interaction, for example, the minimum TTC. The objective of this article is to investigate the impact of motion prediction methods on safety indicators and the interpretation methods of these indicators for safety diagnosis. A framework is proposed to predict road users’ future positions depending on different extrapolation hypotheses: kinematic methods such as constant velocity and motion pattern matching learnt from the observed trajectories. Two interpretation approaches of safety indicators are presented: (a) aggregated safety indicator distributions and (b) indicator profile classification. The framework is applied to the safety diagnosis of left-turn and opposite-direction interactions at a signalized intersection in Montreal, Canada. The results show that motion pattern matching is able to compute the safety indicators earlier and to provide a larger number of measurements.