Predicting Changes in Driving Safety Performance on an Individualized Level Under Naturalistic Driving Conditions

Transportation incidents remain a pressing public safety issue in the United States and throughout the world, despite significant advancements in vehicle safety technologies. The National Highway Traffic Safety Administration (NHTSA) estimates that about 20% of all crashes are fatigue-related, and as such has begun an initiative to reduce drowsy and distracted driving. Of particular interest are commercial truck drivers.

In order to reduce the likelihood of incidents, it is important to understand the factors that affect driver safety performance in order to predict future changes in performance. The goal of this project is to examine how driver safety performance varies by location, time of day, hours on duty, and/or driver workload and to model the rate of change in performance to predict hazardous behaviors.

To meet the overall goal, the following tasks will be completed:

    1) model input parameters for characterizing workload: tasks performed, cognitive load, miles driver, road locations, driving characteristics;

    2) quantify changes in driving performance based on mirror checks and system alerts and evaluate these changes with respect to gold standard guidelines; and

    3) investigate data-driven modeling approaches for driving safety performance prediction, including structural analysis and machine learning approaches.

Data: The research approach makes use of data collected from a fleet of over 100 trucks, with over 1000 hours and 20,000 miles of driving, using the Maven Co-Pilot technology. This data collection tool is a hands-free Bluetooth headset that tracks head motion in each direction at a rate of 50 Hz, transmits this data to the driver’s cell phone, and logs the GPS location to each data point. The system detects driver alertness and Federal Motor Carrier Safety Administration guideline compliance, along with distracted driving and micro-sleeps.