Defining and screening crash surrogate events using naturalistic driving data

Naturalistic driving studies provide an excellent opportunity to better understand crash causality and to supplement crash observations with a much larger number of near crash events. The goal of this research is the development of a set of diagnostic procedures to define, screen, and identify crash and near crash events that can be used in enhanced safety analyses. A way to better understand crash occurrence and identify potential countermeasures to improve safety is to learn from and use near crash events, particularly those near crashes that have a common etiology to crash outcomes. This paper demonstrates that a multi-stage modeling framework can be used to search through naturalistic driving data, extracting statistically similar crashes and near crashes. The procedure is tested using data from the VTTI 100-car study for road departure events. A total of 63 events are included in this application. While the sample size is limited in this empirical study, the authors believe the procedure is ready for testing in other applications.

► A multi-stage framework is developed to search through naturalistic driving data.
► We develop a set of diagnostic procedures to define, screen, and identify surrogate events.
► Surrogate events can be used to identify potential countermeasures to improve safety.