Real-Time Driver Vigilance Monitoring System

Project Overview

We developed a real-time driver vigilance monitoring system and evaluated it in an on-road study with 45 drivers. We compared automated detection with observer ratings and driver self-reports to see how often the methods agreed on early vigilance decline. We also explored which physiological and behavioral signals accompanied those changes.

Background

Maintaining vigilance during prolonged driving can be challenging, particularly when automation reduces how actively drivers engage with the task. Drivers may have difficulty recognizing early changes in their own alertness. A monitoring system could help identify those changes before they become more pronounced.

Research Questions

RQ1. How consistently do automated monitoring, observer ratings, and self-reported assessments identify early vigilance decline?

RQ2. Which physiological and behavioral indicators are associated with early vigilance decline during real-world driving?

Research Approach

We developed a camera-based PERCLOS system that uses facial landmarks to estimate the percentage of time a driver’s eyes are closed and flag potential vigilance decline in real time. We evaluated the system on the road under both manual and Level 2 assisted driving conditions.

We compared three ways to assess driver vigilance.

• Automated monitoring. PERCLOS-based eyelid closure detection.
• Human observation. Ratings by trained observers.
• Self-report. Driver ratings on the Karolinska Sleepiness Scale (KSS).

We also collected eye-tracking, head movement, cardiovascular, and driving performance data to explore complementary indicators of early vigilance decline.

Key Insights

01. Self-reports showed weak agreement with objective assessments. Drivers’ ratings alone may miss early changes in vigilance.

02. Head movement and blink duration showed potential as complementary indicators. Both were associated with episodes of vigilance decline identified through objective assessment.

03. On-road evaluation exposed practical challenges. Sensor noise and variability in driver behavior matter when evaluating monitoring systems in realistic settings.

My Contributions

• System Development. Contributed to developing and implementing the real-time PERCLOS monitoring system.
• Experimental Research. Contributed to evaluating vigilance in manual and Level 2 assisted driving.
• Multimodal Data Analysis. Analyzed physiological and behavioral measurements to investigate early indicators of vigilance decline.
• Research Insights. Interpreted findings to inform the design and evaluation of future driver monitoring systems.

Takeaway

Combining behavioral monitoring with human factors assessments can give designers a fuller picture of early vigilance decline. Where the methods disagree, that disagreement can guide what to measure and validate in future adaptive systems.