An Identification Method for Road Hypnosis Based on Human EEG Data

Bin Wang, Jingheng Wang, Xiaoyuan Wang, Longfei Chen, Han Zhang, Chenyang Jiao, Gang Wang, Kai FengSensors 2024 · DOI

TL;DR

This paper introduces a novel method for identifying "road hypnosis," a dangerous state of reduced awareness in drivers, using electroencephalography (EEG) data. By employing the EEGNet deep learning model, they achieve a high accuracy of 93.01% in detecting this state, offering a promising avenue for enhancing intelligent vehicle safety systems.

What Problem Does This Paper Address?

Road hypnosis, often described as "sleeping with the eyes open," is a phenomenon where drivers enter a state of reduced attention and awareness during prolonged, monotonous driving. This can lead to severe accidents, with personal factors like fatigue and distraction contributing to a significant portion of traffic fatalities. While external signs of road hypnosis are difficult to observe reliably, internal physiological states are key indicators. Existing methods for identifying driver states often rely on external cues or less direct physiological measures. This paper aims to bridge this gap by using electroencephalography (EEG), a direct measure of brain activity, to accurately identify road hypnosis, a state that has been historically understudied using such internal metrics.

Methods

The researchers propose an identification method for road hypnosis based on human EEG data. The process involves several key steps:

  • Data Collection: EEG data was collected from participants during both real-world vehicle driving experiments and virtual driving simulations. These experiments were designed to induce monotonous driving conditions, mimicking environments where road hypnosis is likely to occur (e.g., tunnels, highways).
  • Data Preprocessing: Raw EEG data underwent extensive preprocessing. This included:
  • Data Screening: Manual review and exclusion of data from participants exhibiting signs of fatigue or improper posture.
  • Electrode Location and Re-referencing: Standardized electrode placement and a bilateral mastoid average reference were used.
  • Filtering: Low-pass (30 Hz), high-pass (0.1 Hz), and notch (40 Hz) filters were applied to clean the EEG signals and remove artifacts.
  • Independent Component Analysis (ICA): Used to separate independent brain signals from noise and artifacts, with specific components identified and excluded if they represented artifacts (e.g., eye movement, ECG interference).
  • Epoch Processing: EEG data was segmented into fixed-length epochs (0.2s to 1s) aligned with specific events (road hypnosis or normal driving).
  • Feature Extraction: Power Spectral Density (PSD) analysis was employed to transform the time-domain EEG signals into the frequency domain, extracting key frequency components indicative of brain states.
  • Model Training and Identification: Three neural network architectures were used to build identification models:
  • EEGNet: A specialized convolutional neural network designed for EEG signal analysis.
  • Recurrent Neural Network (RNN): Utilized for its ability to handle sequential data.
  • Long Short-Term Memory (LSTM): An advanced RNN variant designed to capture long-range temporal dependencies. These models were trained and validated using the preprocessed EEG data.

Datasets

The study utilized two primary datasets:

  • Vehicle Driving Experiment Dataset: Collected from 41 participants during real-world driving in monotonous environments. This dataset contained 6063 event-related potentials classified as Road Hypnosis (RH) and 4183 classified as Normal Driving (ND).
  • Virtual Driving Experiment Dataset: Collected from 42 participants in a simulated driving environment. This dataset contained 5182 RH and 7245 ND event-related potentials.

The data was split into 60% for training, 20% for validation, and 20% for testing to ensure robust model evaluation.

Key Insights and Ideas

  • EEG as a Gold Standard for Internal State Monitoring: The paper strongly advocates for EEG as a reliable, non-invasive method for assessing internal driver states, particularly for phenomena like road hypnosis that lack obvious external markers.
  • EEGNet's Superiority for Temporal EEG Features: The study highlights that EEGNet, a network specifically designed for EEG, outperforms general-purpose RNN and LSTM models in capturing the temporal and spectral characteristics crucial for identifying road hypnosis. This suggests that architectural choices tailored to the data modality are critical.
  • Monotony as a Key Inducer: Both real-world and virtual driving experiments were designed to induce monotony, underscoring the environmental factors that contribute to road hypnosis. The virtual environment proved more effective at inducing the state, likely due to controlled conditions.
  • Importance of Preprocessing: The extensive preprocessing pipeline, including ICA for artifact removal, was crucial for extracting clean and relevant EEG features, demonstrating that raw EEG data is insufficient for accurate analysis.

Key Results

The study found that the EEGNet model achieved the highest accuracy of 93.01% in identifying road hypnosis.

  • Virtual Driving Experiments:
    • EEGNet: Accuracy 95.18%, Sensitivity 94.90%, Specificity 90.57%, Geometric Mean 92.69%
    • RNN: Accuracy 89.39%, Sensitivity 90.28%, Specificity 88.51%, Geometric Mean 86.73%
    • LSTM: Accuracy 86.90%, Sensitivity 87.86%, Specificity 81.78%, Geometric Mean 84.76%
  • Vehicle Driving Experiments:
    • EEGNet: Accuracy 92.98%, Sensitivity 94.52%, Specificity 90.89%, Geometric Mean 89.58%
    • RNN: Accuracy 89.88%, Sensitivity 90.33%, Specificity 86.32%, Geometric Mean 87.86%
    • LSTM: Accuracy 88.02%, Sensitivity 89.88%, Specificity 82.89%, Geometric Mean 84.76%

EEGNet consistently outperformed both RNN and LSTM across all evaluated metrics (accuracy, sensitivity, specificity, and geometric mean) in both experimental settings. The virtual driving experiments generally yielded slightly higher performance metrics, likely due to the more consistent induction of the road hypnosis state.

Limitations and Future Work

  • Subjectivity in ICA: The authors acknowledge that the quality of ICA results can depend on researcher judgment, potentially introducing subjective bias in artifact identification. Future work could explore machine learning methods for automated artifact component identification.
  • Data Representativeness: While vehicle driving data is more representative, the accuracy of models built from it was slightly lower than from virtual driving data. This could be due to the larger volume of data and varying degrees of road hypnosis induction in real-world environments.
  • Real-time Application: While the study demonstrates high accuracy, further research is needed to optimize the system for real-time implementation in vehicles, considering computational constraints.
  • Broader Applicability: Future work could explore the generalizability of this method to other states of driver impairment or fatigue and investigate the influence of diverse driving conditions and individual differences.

Summary updated

Abstract

The driver in road hypnosis has not only some external characteristics, but also some internal characteristics. External features have obvious manifestations and can be directly observed. Internal features do not have obvious manifestations and cannot be directly observed. They need to be measured with specific instruments. Electroencephalography (EEG), as an internal feature of drivers, is the golden parameter for drivers' life identification. EEG is of great significance for the identification of road hypnosis. An identification method for road hypnosis based on human EEG data is proposed in this paper. EEG data on drivers in road hypnosis can be collected through vehicle driving experiments and virtual driving experiments. The collected data are preprocessed with the PSD (power spectral density) method, and EEG characteristics are extracted. The neural networks EEGNet, RNN, and LSTM are used to train the road hypnosis identification model. It is shown from the results that the model based on EEGNet has the best performance in terms of identification for road hypnosis, with an accuracy of 93.01%. The effectiveness and accuracy of the identification for road hypnosis are improved in this study. The essential characteristics for road hypnosis are also revealed. This is of great significance for improving the safety level of intelligent vehicles and reducing the number of traffic accidents caused by road hypnosis.
BibTeX
@inproceedings{wang2024identification,
  author = {Bin Wang and Jingheng Wang and Xiaoyuan Wang and Longfei Chen and Han Zhang and Chenyang Jiao and Gang Wang and Kai Feng},
  title = {{An Identification Method for Road Hypnosis Based on Human EEG Data}},
  booktitle = {Sensors},
  year = {2024},
  doi = {10.3390/s24134392},
}
Figure