Adaptive Individual-Level Cognitive Driving Anomaly Detection Model Exclusively Using BSMs

Abstract

Detecting drivers’ cognitive states poses a substantial challenge. In this context, cognitive driving anomalies have generally been regarded as stochastic disturbances. To the best of the author’s knowledge, existing safety studies in the realm of human Driving Anomaly Detection (DAD) utilizing vehicle trajectories have predominantly been conducted at an aggregate level, relying on data aggregated from multiple drivers or vehicles. However, to gain a more nuanced understanding of driving behavior at the individual level, a more detailed and granular approach is essential. To bridge this gap, we developed a Data Anomaly Detection (DAD) model designed to assess a driver’s cognitive abnormal driving status at the individual level, relying solely on Basic Safety Message (BSM) data. Our DAD model comprises both online and offline components, each of which analyzes historical and real-time Basic Safety Messages (BSMs) sourced from connected vehicles (CVs). The training data for the DAD model consist of historical BSMs collected from a specific CV over the course of a month, while the testing data comprise real-time BSMs collected at the scene. By shifting our focus from aggregate-level analysis to individual-level analysis, we believe that the DAD model can significantly contribute to a more comprehensive comprehension of driving behavior. Furthermore, when combined with a Conflict Identification (CIM) model, the DAD model has the potential to enhance the effectiveness of Advanced Driver Assistance Systems (ADAS), particularly in terms of crash avoidance capabilities. It is important to note that this paper is part of our broader research initiative titled “Automatic Safety Diagnosis in the Connected Vehicle Environment”, which has received funding from the Southeastern Transportation Research, Innovation, Development, and Education Center.