⚡ electrical engineering

From Raw Gaze to Meaningful Features: Assessment of Visual Behavior in Driving Simulator

This study demonstrates that a single-sensor, multiparameter eye-tracking framework effectively characterizes driver visual behavior under reduced visibility by revealing significant changes in gaze stability and blinking patterns, thereby providing a practical foundation for developing adaptive driver monitoring systems.

Original authors: Smilja Stokanović, Jaka Sodnik, Nadica Miljković

Published 2026-03-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Smilja Stokanović, Jaka Sodnik, Nadica Miljković

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

🚗 The Big Idea: How Our Eyes Drive When the World Gets Foggy

Imagine you are driving a car. Usually, your eyes are like a security camera that is constantly scanning the road, the mirrors, and the dashboard. It's a busy job! But what happens when a thick fog rolls in? You can't see as far, so your brain has to work harder to figure out where you are going.

This study asked a simple question: How does our eye behavior change when visibility drops?

The researchers didn't just look at where people looked; they looked at how they looked. They used a special pair of glasses (an eye tracker) to record the movements of 24 drivers in a driving simulator. They compared three scenarios:

The "Chill" Mode (Baseline): Just sitting still, relaxing.
The "Normal" Drive: Driving on a clear highway.
The "Foggy" Drive: Driving in thick, scary fog.

They analyzed three main things: Saccades (quick eye jumps), BCEA (how much your eyes wander), and Blinks (how often you close your eyes).

🔍 The Three "Eye Superpowers" They Measured

To make sense of the data, the researchers broke eye behavior down into three categories. Think of it like analyzing a soccer player's performance:

1. Saccades: The "Eye Jumps"

What it is: Your eyes don't move smoothly; they jump from one spot to another in tiny, lightning-fast bursts. These are called saccades.
The Finding: When the drivers were just sitting still, their eyes made many small, random jumps. But when they started driving (especially in fog), they made fewer jumps, but the jumps were bigger.
The Analogy: Imagine a hummingbird hovering over a flower (sitting still). It flutters its wings everywhere. Now, imagine a hawk spotting a mouse in the fog. It stops fluttering and makes one big, decisive dive. The drivers stopped "fluttering" and started making "dive" moves to grab critical information from the fog.

2. BCEA: The "Gaze Wobble"

What it is: This stands for Bivariate Contour Ellipse Area. Sounds complicated, right? Let's call it the "Wobble Zone." It measures how much your eyes drift around a specific spot.
The Finding: When sitting still, your eyes wander a lot (a big Wobble Zone). When driving, especially in fog, your eyes became much more stable. They locked onto the road and stopped wandering.
The Analogy: Think of a toddler playing with a ball. When they are bored (sitting still), they toss the ball everywhere in the room (big wobble). But when they are playing a serious game of catch (driving in fog), they keep the ball in a tight, focused circle. The fog made the drivers' eyes "lock on" tighter.

3. Blinks: The "Strategic Pause"

What it is: How often and how long you blink.
The Finding: This was the most surprising part. When people are tired, they blink a lot and keep their eyes closed longer. But when these drivers were in the fog, they blinked faster and more frequently, but for a shorter time.
The Analogy: Imagine you are trying to read a sign in a strong wind. You don't want to close your eyes for a long time because you might miss something important. So, you do a "rapid-fire" blink—blink-blink-blink—just to refresh your eyes without losing the view. The drivers were essentially "strategically suppressing" their blinks to keep the foggy road in sight.

🆕 The New "Secret Weapon": Guzik's Index (GI)

The researchers invented a new math trick called Guzik's Index (GI).

What it does: It measures if your eyes are wobbling more up-and-down or side-to-side.
The Finding: In the fog, the drivers' eyes became very symmetrical and focused. The GI helped prove that the drivers weren't just "looking around"; they were adapting their gaze to be perfectly efficient for the foggy conditions.
The Analogy: Think of a flashlight. In a normal room, you might swing the light around the whole room. In a dark tunnel (fog), you point the beam straight ahead and hold it steady. The GI measures how "straight" that beam is.

🧠 What Does This Mean for the Future?

The study concludes that one single sensor (the eye tracker) is enough to tell if a driver is struggling with poor visibility.

Why is this cool?
Currently, cars have systems that tell you if you are tired (by checking if your eyes are heavy). But this study shows that eye trackers can also tell you if you are stressed by the weather.

The Future Application:
Imagine a "Smart Car Dashboard" that talks to your eyes:

Scenario: You are driving in heavy fog. The car's eye tracker notices your eyes are making fewer, bigger jumps and blinking very quickly.
The Car's Reaction: "Ah, the driver is in high-stress mode. I will simplify the dashboard. I'll make the text bigger, hide the radio ads, and maybe even slow down the car automatically."

🏁 The Takeaway

When the world gets foggy, our eyes don't just panic; they adapt. They stop wandering, they make bigger jumps to find clues, and they blink faster to stay alert. By understanding these "eye secrets," we can build cars that are smarter, safer, and more helpful when the weather turns bad.

1. Problem Statement

Driving under reduced visibility conditions, such as fog, presents a significant safety risk due to the increased cognitive and visual workload required to maintain situational awareness. While eye tracking is a standard tool for monitoring driver attention, current research and driver monitoring systems (DMS) often rely on unimodal approaches (measuring only saccades, fixations, or blink rates separately) or focus primarily on clear visibility conditions.

There is a critical gap in understanding how oculomotor behavior adapts specifically to reduced visibility. Existing single-metric systems often fail to capture the complex, simultaneous changes in gaze stability, scanning patterns, and blinking behavior that occur when drivers suppress blinks to maintain visual input and alter their gaze dispersion to cope with environmental strain. The study aims to determine if a single-sensor, multiparameter eye-tracking approach can effectively characterize these adaptations to inform the design of adaptive Human-Computer Interfaces (HCI) and Advanced Driver-Assistance Systems (ADAS).

2. Methodology

Experimental Setup:

Participants: 24 valid participants (out of an initial 28) with normal or corrected-to-normal vision and valid driver's licenses.
Environment: A motion-based driving simulator at the University of Ljubljana, featuring a 145° field of view across three HD screens.
Conditions: Three distinct phases were analyzed:
1. Baseline: 15 minutes of sitting still and relaxing.
2. Ride: ~10 minutes of highway driving under normal visibility.
3. Fog: ~2 minutes of highway driving with dense fog (reduced visibility).
Data Acquisition: Eye movements were recorded using a Tobii Pro Glasses 3 eye tracker at 100 Hz.

Data Processing & Feature Extraction:

Preprocessing: Data was smoothed using a 3rd-order Savitzky-Golay filter (110 ms window) to reduce noise. Missing data was handled via median imputation or exclusion.
Algorithm: An adaptive I-VT (Identification - Velocity Threshold) algorithm was used for saccade detection, with parameters optimized to minimize misdetections based on physiological duration limits (10–100 ms).
Feature Categories (31 Parameters Total):
1. Saccade Parameters (SP): Amplitude, peak velocity, duration, number of saccades, and fixation duration.
2. Bivariate Contour Ellipse Area (BCEA) Parameters: A spatial measure of gaze dispersion including:
  - Standard deviations of horizontal/vertical gaze.
  - Correlation coefficient ( $\rho$ ).
  - Number of Preferred Retinal Loci (PRL).
  - Guzik's Index (GI): A novel metric introduced in this study to quantify fixation asymmetry along the major BCEA axis (ratio of SD parallel to perpendicular to the major axis).
  - Mean Squared Error (MSE) and Entropy measures.
3. Blinking Parameters (BP): Blink duration, blink rate, and number of blinks per minute.

Statistical Analysis:

Repeated measures ANOVA (parametric) and Friedman tests (non-parametric) were used to compare the three conditions.
Post-hoc pairwise comparisons were performed with Benjamini-Hochberg False Discovery Rate (FDR) correction.
Effect sizes were calculated using Cohen's $d$ and Cliff's $\delta$ .

3. Key Contributions

Multiparameter Framework: The study proposes a comprehensive framework integrating saccade dynamics, spatial gaze dispersion (BCEA), and blink patterns into a single analytical model, moving beyond single-metric limitations.
Introduction of Guzik's Index (GI) for Driving: Adapting a metric originally used in ECG analysis, the authors introduced GI to quantify gaze asymmetry. This provides a new dimension to analyze how drivers distribute their visual attention relative to the primary axis of their gaze ellipse.
First Application of BCEA in Driving Simulators: The study applies BCEA metrics to characterize driver visual attention stability in dynamic driving scenarios, demonstrating its sensitivity to visibility changes.
Open Science: The authors released the anonymized dataset and the full source code (including I-VT implementation, BCEA calculation, and entropy measures) on GitHub and Zenodo under open licenses.

4. Key Results

Saccade Behavior:

Amplitude: Saccade amplitudes were significantly larger during driving (Ride and Fog) compared to Baseline, suggesting drivers make broader gaze shifts to scan the environment under load.
Fixation: Fixation durations were significantly shorter during driving conditions, indicating rapid visual sampling.
Frequency: The total number of saccades decreased significantly during driving compared to Baseline.
Visibility Impact: Saccade features showed minimal statistically significant differences between Ride and Fog (only 1 of 13 parameters changed), suggesting saccade dynamics stabilize once driving begins, regardless of fog.

Gaze Stability (BCEA):

Dispersion: BCEA was significantly larger during Baseline (resting) than during driving. During driving, gaze became more focused.
Fog Effect: Gaze dispersion was lowest during the Fog condition compared to Ride, indicating that reduced visibility forces drivers to narrow their visual field and stabilize their gaze more intensely.
GI & MSE: Guzik's Index was significantly higher in Baseline, indicating greater asymmetry during rest. During driving, GI decreased, and MSE distinguished Ride from Fog, confirming that gaze becomes more symmetric and stable under high-demand conditions.

Blinking Behavior:

Suppression Pattern: Blink duration significantly decreased (shorter blinks) during driving, while blink rate and frequency increased compared to Baseline.
Fog vs. Ride: Blink rate was slightly higher in Fog than Ride, but the most distinct change was the shift from long, variable blinks in Baseline to short, consistent blinks during driving. This suggests a strategic suppression of long blinks to maintain visual continuity in fog.

5. Significance and Implications

Driver Monitoring Systems (DMS): The study validates that a single sensor (eye tracker) combined with a multiparameter approach is sufficient to detect complex states of visual strain and reduced visibility. It proves that relying on a single metric (e.g., just blink rate) is insufficient; the combination of reduced gaze dispersion (BCEA), altered saccade patterns, and blink suppression provides a robust signature of high workload.
Adaptive HCI Design: The findings suggest that automotive interfaces should be adaptive. When the system detects the specific "fog signature" (narrowed gaze, high blink rate, short duration), it should:
- Simplify display information.
- Increase text size.
- Delay non-critical secondary alerts to avoid cognitive overload.
Safety Evaluation: The introduction of GI and BCEA offers new, sensitive metrics for evaluating driver situation awareness and visual stability in challenging weather, potentially leading to earlier detection of driver fatigue or distraction in real-world scenarios.

In conclusion, the paper demonstrates that drivers dynamically modulate their oculomotor behavior to cope with environmental challenges. By quantifying these adaptations through a novel multiparameter framework, the study provides a technical foundation for next-generation, context-aware driver monitoring systems.