EEG-based Driver Fatigue Detection with Eye-tracking Guided Weak Supervision
DOI: 10.71052/srb2024/aafr1341
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This study addresses the challenges of limited interpretability, subjective labeling, and insufficient temporal modeling in existing EEG-based driver fatigue detection systems. While EEG signals provide direct neurophysiological insights, traditional methods often rely on handcrafted features that are sensitive to noise and fail to capture the progressive nature of fatigue. To overcome these limitations, the authors propose a novel framework that integrates EEG band-ratio features with eye-tracking-derived weak supervision labels. The approach aims to enhance both the accuracy and interpretability of fatigue state classification by leveraging automated, objective labeling and deep learning architectures capable of modeling long-term temporal dependencies. The experimental design involved 20 participants engaged in a 120-minute continuous simulated driving task using Euro Truck Simulator 2, configured to follow a 24-hour day-night cycle to induce natural fatigue progression. Data were collected using Tobii Pro Glasses 3 for eye-tracking and an eego mylab portable EEG system. Preprocessing included artifact removal and band-pass filtering into five canonical frequency bands (Delta, Theta, Alpha, Beta, Gamma). Five physiologically meaningful band-ratio features ($\alpha/\beta$, $\theta/\beta$, $\theta/\alpha$, $\delta/\theta$, $\gamma/\beta$) were extracted to characterize relative neural activity changes. Weak supervision labels for three fatigue states (alert, mild, severe) were automatically generated using a dual-threshold strategy based on the PERCLOS metric derived from pupil diameter data. The core model, a Stacked LSTM Attention Network (SLAN), processes these feature sequences using a two-layer LSTM to capture temporal dynamics and an attention mechanism to weight critical time segments. Results demonstrate that the proposed SLAN model significantly outperforms baseline models, including ESTCNN, EEGNet, and Interpretable CNN. The SLAN achieved an accuracy of 97.45%, a precision of 91.22%, a recall of 88.88%, and an F1-score of 90.01, while reducing the miss rate to 11.12%. Comparative analysis revealed that fusing all five band-ratio features yielded superior performance compared to any single feature, improving the F1-score by approximately 3.3% over the best individual feature ($\delta/\theta$). Attention weight visualization indicated that the model prioritizes recent temporal segments, aligning with the cumulative nature of fatigue. This suggests the model effectively identifies dynamic temporal patterns in EEG activity associated with fatigue progression. The significance of this work lies in its effective integration of weakly supervised eye-tracking labels with deep EEG feature learning, reducing reliance on subjective manual annotation. The SLAN architecture enhances interpretability by highlighting discriminative temporal segments, offering insights into the neural mechanisms of fatigue. These findings provide a robust foundation for developing continuous, adaptive driver monitoring systems in intelligent cockpit applications, potentially improving road safety by minimizing missed fatigue detections.
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | Crossref | — | — | 1 | 2026-08-09 |
| archive | success | canonical_url | — | — | 1 | 2026-08-09 |
| extract | success | pdftotext | — | — | 125 | 2026-08-10 |
| clean | success | clean | — | — | 2 | 2026-08-10 |
| chunk | success | chunk | — | — | 2 | 2026-08-10 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 2 | 2026-08-10 |
| promote | success | — | — | — | 1 | 2026-08-09 |
| summarize | success | llm | qwen3.6-27b-nvidia | summ-v5 | 123 | 2026-08-10 |
| tag | success | vector_similarity | — | — | 17 | 2026-08-11 |
| verify | partial | — | — | — | 2 | 2026-08-10 |
Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified_with_issues.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Empirical Findings: physiological data
- Methodological Resource: tool software