EEG-based emergency braking intention detection during simulated driving

Liang, Xinbin; Yu, Yang; Liu, Yadong; Liu, Kaixuan; Liu, Yaru; Zhou, Zongtan · 2023 · Crossref

DOI: 10.1186/s12938-023-01129-4

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 limitations of existing electroencephalogram (EEG)-based driver monitoring systems, which primarily distinguish emergency braking from normal driving but fail to differentiate between emergency and normal braking. The authors argue that this distinction is critical for safety, as misidentifying normal braking as emergency could lead to unnecessary interventions, while missing emergency braking could result in collisions. Additionally, previous research relied heavily on manually extracted features and traditional machine learning algorithms, potentially overlooking complex neural patterns. To overcome these issues, the researchers proposed a novel detection strategy using raw EEG signals as input for traditional, Riemannian geometry-based, and deep learning-based classification methods. The experiment was conducted on a simulated driving platform with 10 subjects across three scenarios: normal driving, normal braking, and emergency braking. In emergency braking trials, subjects responded to external cues (lead vehicle brake lights), whereas normal braking was performed spontaneously. EEG signals were recorded and preprocessed with simple filtering and baseline correction. The study compared six algorithms: Common Spatial Pattern with Linear Discriminant Analysis (CSP + LDA), xDAWN + LDA, Riemannian Minimum Distance to Mean (RMDM), Tangent Space with Logistic Regression (TS + LR), EEGNet, and ShallowConvNet. Performance was evaluated using the Area Under the Curve (AUC) and F1 score, with data augmentation applied to some models to enhance deep learning performance. Results indicated significant differences in EEG potentials between emergency and normal braking, particularly in the parieto-occipital and frontal-central regions. Emergency braking elicited a pronounced P300-like positive potential shift and larger negative shifts associated with movement planning, whereas normal braking showed earlier, smaller amplitude changes. Classification results demonstrated that Riemannian geometry-based (TS + LR) and deep learning-based (EEGNet) methods outperformed traditional approaches. At 200 ms before braking onset, the EEGNet algorithm with data augmentation achieved an AUC of 0.94 and an F1 score of 0.65 for distinguishing emergency braking from normal driving. For distinguishing emergency from normal braking, it achieved an AUC of 0.91 and an F1 score of 0.85. Conversely, distinguishing normal braking from normal driving remained difficult, with optimal F1 scores below 0.20. Data augmentation significantly improved the performance of deep learning models, making them comparable to or better than Riemannian methods. The study concludes that detecting emergency braking intention from both normal driving and normal braking using raw EEG signals is feasible. The ability to identify emergency braking intentions 200 ms before physical action could allow automatic braking systems to activate earlier, potentially reducing braking distance by approximately 5 meters at 90 km/h. This framework supports human-vehicle co-driving by enabling proactive safety interventions. The authors note that while simulated environments provide safety, future work should address real-world variability and the computational efficiency of different algorithms for online implementation.

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.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms

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: computational model