Classification of Driver Cognitive Load in Conditionally Automated Driving: Utilizing Electrocardiography and Eye Tracking
DOI: 10.1177/03611981241252797
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 challenge of monitoring driver cognitive load in conditionally automated vehicles (SAE Level 3), where drivers may engage in non-driving tasks that impair their ability to safely take over control. While physiological measures are promising for this estimation, existing methods rely on manual extraction of low-level features from signals like electrocardiograms (ECG), which often results in information loss and poor adaptability to dynamic conditions. The authors propose a novel deep learning approach that utilizes ECG spectrograms to automatically capture high-level features, aiming to improve classification accuracy and practical feasibility for vehicle integration. The researchers developed a deep learning framework called SE-ECG Net, which combines a convolution module with a Squeeze-and-Excitation (SENet) attention mechanism to identify cognition-related features within 2D ECG spectrograms. These spectrograms were generated by applying a short-time Fourier Transform to preprocessed ECG signals, which underwent downsampling, denoising, R-peak detection, and segmentation into 60-second windows. The model was evaluated on two datasets: a public dataset involving 87 subjects performing verbal-cognitive tasks in a driving simulator, and an unpublished dataset with 42 participants. Performance was assessed using both within-subject and across-subjects data partitions via 10-fold cross-validation, comparing the proposed method against baseline models (VGG Net, ResNet, and CNNs) and traditional hand-crafted feature extraction methods. Results demonstrated that spectrogram-based models significantly outperformed those using hand-crafted features. The proposed SE-ECG Net achieved the highest accuracy, reaching 96.76% in within-subject evaluations and 71.50% in across-subject evaluations on the primary dataset. The inclusion of the SENet attention module contributed to superior performance compared to standard convolutional networks, as it effectively weighted important channels in the spectrogram. However, a notable performance drop occurred in across-subject evaluations, indicating that the models struggled to fully generalize individual-independent features across different drivers. Confusion matrices revealed that models were generally better at identifying low cognitive load than high cognitive load, potentially due to signal ceiling effects. The study concludes that using ECG spectrograms with deep learning is a feasible and effective method for estimating driver cognitive load, offering advantages over traditional feature extraction by preserving more information. The approach is practically valuable because ECG sensors can be integrated into steering wheels, providing a non-intrusive alternative to eye-tracking or EEG. Despite the promising results, the authors note limitations regarding generalization across individuals and the need for further validation in real-world driving environments. Future work should focus on improving cross-subject performance through techniques like domain generalization and data fusion, as well as addressing sensor reliability and noise reduction for real-time deployment.
Key finding
A deep learning model using ECG spectrograms achieved superior accuracy in classifying driver cognitive load compared to models using hand-crafted features, validating the feasibility of this approach for monitoring drivers in conditionally automated vehicles.
Methodology
simulation_modeling
Sample size: 132
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 | — | — | — | 1 | 2026-05-28 |
| archive | success | canonical_url | — | — | 1 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | skipped | — | — | — | 3 | 2026-06-04 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
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
- Theoretical Contribution: theory or model