Cross-Subject Zero Calibration Driver’s Drowsiness Detection: Exploring Spatiotemporal Image Encoding of EEG Signals for Convolutional Neural Network Classification

Paulo, Joao Ruivo; Pires, Gabriel; Nunes, Urbano J. · 2021 · Crossref

DOI: 10.1109/tnsre.2021.3079505

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of developing a cross-subject, zero-calibration system for detecting driver drowsiness using electroencephalography (EEG) signals. The motivation stems from the high incidence of traffic accidents caused by fatigue and the limitations of existing EEG-based systems, which typically require subject-specific calibration due to low signal-to-noise ratios and significant inter-subject variability. The authors propose a methodology that transforms EEG data into spatiotemporal image encodings, enabling the use of Convolutional Neural Networks (CNNs) for classification without individual calibration steps. The study utilizes a public dataset comprising EEG recordings from 27 participants engaged in a simulated sustained-attention driving task. EEG signals were collected via 30 channels at 500 Hz. The authors extracted 3-second pre-event windows, labeling them as "Alert" or "Drowsy" based on reaction times to lane-departure events. Specifically, windows with reaction times below 1.5 times the subject's baseline alert reaction time were labeled "Alert," while those exceeding 2.5 times were labeled "Drowsy." The methodology involves band-pass filtering the signals into theta, alpha, and beta sub-bands, computing cumulative band powers and an attention metric over 50 ms frames, and encoding these features into image-like structures. Two encoding techniques were explored: Recurrence Plots (RP), which capture temporal and spatiotemporal recurrences, and Gramian Angular Fields (GAF), which preserve temporal correlations by transforming time series into polar coordinates. These images served as inputs for a CNN with a single convolutional layer. Neurophysiological analysis revealed that theta and alpha bands exhibited the most discriminative power across frontal, central, and parietal regions, with beta bands also showing relevance. The experimental results, evaluated using leave-one-out cross-validation, demonstrated a superior balanced accuracy of up to 75.87%. This performance outperforms many existing literature benchmarks for cross-subject drowsiness detection. The study confirms that encoding EEG signals into recurrence plots and gramian angular fields effectively mitigates the issues of non-stationarity and inter-subject variability. The significance of this work lies in its demonstration that deep learning models can achieve robust cross-subject classification without the need for user-specific calibration data. By leveraging spatiotemporal image encodings, the proposed system addresses the practical barriers to deploying EEG-based driver assistance systems, such as the impracticality of lengthy calibration periods. The findings suggest that recurrence plots and gramian angular fields are effective tools for handling the nonlinear and non-stationary nature of EEG signals, paving the way for more accessible and scalable real-time drowsiness detection systems.

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.

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.

• drowsiness detection algorithms
• drowsiness
• microsleep

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