Multimodal Brain–Computer Interface for In-Vehicle Driver Cognitive Load Measurement: Dataset and Baselines

Angkan, Prithila; Behinaein, Behnam; Mahmud, Zunayed; Bhatti, Anubhav; Rodenburg, Dirk; Hungler, Paul; Etemad, Ali · 2024 · Crossref

DOI: 10.1109/tits.2023.3345846

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 introduces CL-Drive, a novel multimodal dataset designed to assess driver cognitive load in real-time driving scenarios. The research addresses critical gaps in existing literature, where cognitive load datasets are often collected in non-driving contexts, rely on sparse subjective ratings, or lack the integration of Brain-Computer Interface (BCI) signals with auxiliary physiological data. High cognitive load is a significant contributor to road accidents, yet current intelligent vehicle systems lack robust methods for continuous, accurate monitoring. The authors aim to provide a comprehensive resource that combines Electroencephalogram (EEG) signals with Electrocardiography (ECG), Electrodermal Activity (EDA), and eye-tracking data, paired with dense, frequent ground-truth labels to facilitate the development of automated alarm systems. The data was collected from 21 subjects using an immersive vehicle simulator equipped with motion feedback and high-resolution visual displays. Participants performed nine distinct driving scenarios, each lasting three minutes and designed to induce progressively higher levels of cognitive complexity, ranging from simple highway driving to complex maneuvers like three-point turns and narrow alley navigation. Physiological signals were recorded using a Muse S1 EEG headband (256 Hz), Shimmer2 sensors for ECG (512 Hz) and EDA (128 Hz), and a Tobii eye tracker (50 Hz). Crucially, unlike prior studies that collected subjective ratings only at task completion, participants reported their cognitive load every 10 seconds using the 9-point Performance Assessment of Anxiety and Stress (PAAS) scale. This frequent self-reporting provides dense ground-truth labels for model training. The study presents benchmark classification results for both binary and ternary cognitive load distributions using various machine learning and deep learning models. The authors evaluated performance using 10-fold cross-validation and leave-one-subject-out (LOSO) criteria, testing models on both hand-crafted features and raw signal data. The results demonstrate that cognitive load induced by driving can be measured with reasonable accuracy using the proposed multimodal approach. The integration of EEG with auxiliary wearable signals proves effective in capturing the neural and physiological correlates of mental workload during dynamic driving tasks. The significance of this work lies in the release of CL-Drive as a public dataset, which is the first to combine driver cognitive load ratings with multimodal bio-signals in a driving context. By providing dense, frequent subjective labels and diverse physiological data, the dataset enables more reliable training of machine learning models for real-time cognitive load assessment. This contribution supports the advancement of intelligent vehicle technologies capable of detecting dangerous levels of driver mental workload, potentially reducing accidents caused by cognitive overload. The paper establishes a baseline for future research in multimodal BCI applications for automotive safety.

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.

• workload measurement
• mental demand
• drowsiness detection algorithms
• distraction detection algorithms
• neuro workload indices
• cognitive capacity variation

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