Road safety engineering studies in the past typically involved regression-based crash prediction models relating crash frequency to road design or operational configurations. In addition to this conventional approach there has been an ongoing interest in using microscopic simulation to predict the safety consequences of engineering decisions, similar to how microscopic models that are used to predict operational performances. Historically,a major obstacle to the development of such crash-inclusive simulation models has been a scarcity of data on driver behavior in crash and near-crash conditions. In the recent years however, with more advanced and sophisticated data collection techniques emerging, it has been possible to collectindividual vehicle data under naturalistic driving conditions. In this study we illustrate using vehicle based data to show how trajectory based modeling technique can be implemented to reconstruct crash related events, and in turn estimate the posterior distribution of important event parameters such as braking accelerations, reaction time, and critical headways for a given set of trajectory data. Our results suggest that for a rear ending event it is possible to obtain precise estimates of the interaction between the leading and the following vehicle and identify the critical driving conditions that cause the event. Given sufficiently large samples of crash and near-crash events, this method could be used to compile distributions for these inputs, which could in turn be used to include realistic crash features in a microscopic simulation model.
Abstract