The Classification of Short and Long-term Driving Behavior for an Advanced Driver Assistance System by Analyzing Bidirectional Driving Features

TL;DR

This study develops a rule-based classification method to identify short-term and long-term driving behaviors using bidirectional features, validated on real-world connected vehicle data, to enhance personalized ADAS feedback.

Contribution

It introduces a novel, simple rule-based approach to classify driving behaviors and habits based on bidirectional control features from naturalistic driving data.

Findings

Achieved about 90% accuracy in identifying speeding instances.

Successfully classified driving behaviors into safe and hostile groups.

Validated the model on real-world connected vehicle data.

Abstract

Insight into individual driving behavior and habits is essential in traffic operation, safety, and energy management. With Connected Vehicle (CV) technology aiming to address all three of these, the identification of driving patterns is a necessary component in the design of personalized Advanced Driver Assistance Systems (ADAS) for CVs. Our study aims to address this need by taking a unique approach to analyzing bidirectional (i.e. longitudinal and lateral) control features of drivers, using a simple rule-based classification process to group their driving behaviors and habits. We have analyzed high resolution driving data from the real-world CV-testbed, Safety Pilot Model Deployment, in Ann Arbor, Michigan, to identify diverse driving behavior on freeway, arterial, and ramp road types. Using three vehicular features known as jerk, leading headway, and yaw rate, driving characteristics are classified into two groups (Safe Driving and Hostile Driving) on short-term classification, and drivers habits are categorized into three classes (Calm Driver, Rational Driver, and Aggressive Driver). Proposed classification models are tested on unclassified datasets to validate the model conviction regarding speeding and steep acceleration. Through the proposed method, behavior classification has been successfully identified about 90 percent of speeding and similar level of acute acceleration instances. In addition, our study advances an ADAS interface that interacts with drivers in real-time in order to transform information about driving behaviors and habits into feedback to individual drivers. We propose an adaptive and flexible classification approach to identify both short-term and long-term driving behavior from naturalistic driving data to identify and, eventually, communicate adverse driving behavioral patterns.