Driving-Induced Neurological Biomarkers in an Advanced Driver-Assistance System

Iqram Hussain, Seo Young, Se Jin Park2021 · DOI

TL;DR

This paper explores how electroencephalogram (EEG) signals can serve as neurological biomarkers for driver mental workload in a simulated driving environment. By analyzing EEG data from resting and driving states (city and expressway), the researchers identified specific brainwave patterns that change with cognitive load, paving the way for more sophisticated advanced driver-assistance systems (ADAS).

What Problem Does This Paper Address?

Driving is a complex task demanding significant mental effort, and increased cognitive load can lead to fatigue, drowsiness, and ultimately, accidents. While existing ADAS rely on vehicle dynamics and external sensors, they often lack a direct understanding of the driver's internal cognitive state. This paper addresses the challenge of quantifying a driver's mental workload in real-time using physiological signals, specifically EEG, to develop more proactive and intelligent ADAS that can adapt to the driver's cognitive demands.

Methods

The study employed a multi-faceted approach:

  • Driving Simulation: A high-fidelity driving simulator with a 360-degree view and motion capabilities was used to create realistic driving scenarios.
  • EEG Data Acquisition: A portable, four-channel EEG headset was used to record brain activity from 17 healthy male drivers during resting, city-roadway driving, and expressway driving states. Electrooculography (EOG) and electromyography (EMG) were also used for artifact removal.
  • Signal Pre-processing: Standard techniques like filtering (60 Hz AC noise, band-pass 0.5-44 Hz), downsampling (to 250 Hz), and Independent Component Analysis (ICA) were applied to clean the EEG data and remove artifacts.
  • Feature Extraction: EEG data was analyzed in the frequency domain. Key features extracted included:
  • Spectral Power: Power within delta, theta, alpha, beta, and gamma bands.
  • Derived Ratios: Delta-Alpha Ratio (DAR) and Delta-Theta Ratio (DTR).
  • Asymmetry: Frontal EEG asymmetry, reflecting differences in left and right hemisphere activity.
  • Statistical Analysis: Paired-samples t-tests were used to compare EEG features between resting and driving states, with a significance level of p < 0.05.
  • Machine Learning Classification: Various machine learning models (KNN, Discriminant Analysis, SVM, C5.0, QUEST) were trained and tested to classify neurological states into resting, city-roadway driving, and expressway driving. Feature selection was performed using F-statistics.

Datasets

The study utilized a dataset collected from 17 healthy male drivers performing simulated driving tasks. This dataset, comprising EEG recordings from resting, city-roadway, and expressway driving states, was used for both statistical analysis and training/testing machine learning models. No external datasets were used for evaluation.

Key Insights and Ideas

  • EEG as a Mental Workload Indicator: The study strongly suggests that EEG spectral power, particularly in delta, theta, alpha, and beta bands, along with DAR and DTR, can serve as reliable biomarkers for quantifying mental workload during driving.
  • Driving Scenario Impact: Different driving scenarios (city vs. expressway) elicit distinct neurological responses, with the city-roadway demanding a higher cognitive load, reflected in increased delta and theta power and decreased beta and gamma power.
  • Asymmetry Matters: Frontal EEG asymmetry, especially in theta waves, showed significant differences between driving states, highlighting its potential as a biomarker for cognitive demand.
  • Machine Learning Efficacy: Machine learning models demonstrated high accuracy in distinguishing between resting and driving states, and moderate accuracy in differentiating between the two driving states, indicating the potential for real-time classification.

Key Results

  • Spectral Power Changes:
    • Delta and theta power increased significantly during driving compared to resting.
  • Beta and gamma power decreased significantly during driving compared to resting.
  • Alpha power showed increases in driving states, with frontal alpha showing a more pronounced increase in city-roadway driving.
  • DAR and DTR: These ratios showed strong correlations with driving states and were significant in distinguishing between resting and driving conditions.
  • Machine Learning Performance:
    • Binary Classification (Resting vs. Driving): Near-perfect accuracies (up to 99.79%) were achieved by SVM and other models.
  • Multi-Class Classification (Resting, City, Expressway): Moderate accuracies were observed, with the best performance around 78.92% (SVM) for training and 68.70% for testing. C5.0 model achieved the highest training accuracy (96.46%) but a lower testing accuracy (64.63%).

Limitations and Future Work

  • Dataset Size: The study used a relatively small dataset of 17 drivers, which limits the generalizability of the findings.
  • Simulator vs. Real-World: While a high-fidelity simulator was used, it may not perfectly replicate the complexities and nuances of real-world driving.
  • Limited Brain Regions: The analysis focused primarily on frontal and occipital lobes, with potential for deeper insights from other cortical regions.
  • Future Work: The authors suggest extending the study with a larger, more diverse participant pool to generalize the approach. They also plan to explore other physiological features (e.g., cardiac signals) and investigate driving in more varied traffic conditions. Further research into individual cortical lobe analysis and real-world driving scenarios is also recommended.

Summary updated

BibTeX
@misc{hussain2021driving,
  author = {Iqram Hussain and Seo Young and Se Jin Park},
  title = {{Driving-Induced Neurological Biomarkers in an Advanced Driver-Assistance System}},
  year = {2021},
  doi = {10.3390/s21216985},
}
Figure