Evaluating Driver Features for Cognitive Distraction Detection and Validation in Manual and Level 2 Automated Driving

Area of Science:

• Human-Computer Interaction
• Automotive Engineering
• Cognitive Psychology

Background:

• Real-time driver state monitoring is essential for enhancing road user safety.
• Understanding driver cognitive distraction (CD) is crucial for developing advanced driver-assistance systems (ADAS).

Purpose of the Study:

• To evaluate the impact of feature selection on the detection and validation of driver cognitive distraction (CD).
• To compare the efficacy of different feature sets in manual and Level 2 automated driving scenarios.

Main Methods:

• Twenty-four participants drove a Tesla Model S in manual and Autopilot modes on a highway.
• Cognitive distraction was assessed using the N-back task.
• A random forest algorithm classified CD using either "hand-crafted" glance features (e.g., percent road center) or numerous machine-generated features from a driver monitoring system (DMS).

Main Results:

• In manual driving, a small set of glance features performed comparably to a large set of machine-generated features for CD classification.
• In Level 2 automated driving, both glance and vehicle features showed reduced sensitivity to CD.
• Glance features indicated that misclassifications stemmed from fluctuating cognitive loads and individual differences.

Conclusions:

• Glance metrics are vital for accurately detecting and validating driver cognitive distraction in real-world driving.
• Human factors expertise in feature selection and ground truth validation is valuable for advancing driver monitoring technologies.