Performance of basic kinematic thresholds in the identification of crash and near-crash events within naturalistic driving data

Abstract

Understanding causal factors for traffic safety-critical events (e.g., crashes and near-crashes) is an important step in reducing their frequency and severity. Naturalistic driving data offers unparalleled insight into these factors, but requires identification of situations where crashes are present within large volumes of data. Sensitivity and specificity of these identification approaches are key to minimizing the resources required to validate candidate crash events. This investigation used data from the Second Strategic Highway Research Program Naturalistic Driving Study (SHRP 2 NDS) and the Canada Naturalistic Driving Study (CNDS) to develop and validate different kinematic thresholds that can be used to detect crash events. Results indicate that the sensitivity of many of these approaches can be quite low, but can be improved by selecting particular threshold levels based on detection performance. Additional improvements in these approaches are possible, and may involve leveraging combinations of different detection approaches, including advanced statistical techniques and artificial intelligence approaches, additional parameter modifications, and automation of validation processes.