Driving stability analysis using naturalistic driving data with random matrix theory

Abstract

Driving behavior analysis has diverse applications in intelligent transportation systems (ITSs). The naturalistic driving data potentially contain rich information regarding human drivers’ habits and skills in practical and natural driving conditions. But mining knowledge from them is challenging. In this paper, we propose a novel approach for analyzing driving stability using naturalistic driving data. Our method can extract features, based on the random matrix theory, to reflect the statistical difference between actual driving data and the data that would be generated by a theoretically ideal driver, and thus imply the skillful level of a driver in terms of vehicle control in both longitudinal and lateral directions. The execution of our method on a practical ITS dataset is conducted. Using the extracted features, a driving behavior analysis application that partitions drivers into clusters to identify common driving stability characteristics is demonstrated and discussed.