Driver Behavior Modeling in Critical Situations for Threat Assessment of Intelligent Vehicles

Abstract

Threat assessment (TA) method is a crucial part in the decision-making process of intelligent vehicles (IVs). Probabilistic threat assessment (PTA) method, as a robust TA method, has drawn increasing attention. This paper utilized vehicle event data recorder (EDR) data to model driver behavior in critical situations for PTA. A total of 415 valid EDR samples were extracted and processed, and multivariate Gaussian distribution (MGD) model was employed to construct the probability density function (PDF) of the driver’s evasive maneuvers. The results of the statistical analysis for the obtained data were in accordance with previous studies, and six Gaussian PDFs classified by different critical conditions were obtained. Moreover, the Gaussian mixture model (GMM) was used to improve the PDFs by introducing the driver’s unresponsive samples. This work can provide accurate models of driver behavior in critical situations and contribute to the development of TA for IVs.