Masked EEG Modeling for Driving Intention Prediction

TL;DR

This work tackles driving safety under drowsy conditions by decoding driving intentions from EEG signals. It introduces Masked EEG Modeling (MEM), a self-supervised, transformer-based framework that learns robust, frequency-domain EEG representations through channel- and frequency-masking, and jointly predicts left/right/straight driving maneuvers. ICA-guided analysis highlights central-frontal and parietal involvement, while Welch PSD analysis identifies discriminative frequency bands, contributing both neuroscientific insight and practical robustness to channel loss. MEM achieves high accuracy (e.g., $85.19\%$ in drowsy states) and preserves performance under substantial channel disruption, signaling strong potential for artifact-tolerant, real-world BCI-assisted driving systems. The approach offers a scalable path to anticipatory driver-support mechanisms, integrating neural signals with automated driving systems to align actions with human intent.

Abstract

Driving under drowsy conditions significantly escalates the risk of vehicular accidents. Although recent efforts have focused on using electroencephalography to detect drowsiness, helping prevent accidents caused by driving in such states, seamless human-machine interaction in driving scenarios requires a more versatile EEG-based system. This system should be capable of understanding a driver's intention while demonstrating resilience to artifacts induced by sudden movements. This paper pioneers a novel research direction in BCI-assisted driving, studying the neural patterns related to driving intentions and presenting a novel method for driving intention prediction. In particular, our preliminary analysis of the EEG signal using independent component analysis suggests a close relation between the intention of driving maneuvers and the neural activities in central-frontal and parietal areas. Power spectral density analysis at a group level also reveals a notable distinction among various driving intentions in the frequency domain. To exploit these brain dynamics, we propose a novel Masked EEG Modeling framework for predicting human driving intentions, including the intention for left turning, right turning, and straight proceeding. Extensive experiments, encompassing comprehensive quantitative and qualitative assessments on public dataset, demonstrate the proposed method is proficient in predicting driving intentions across various vigilance states. Specifically, our model attains an accuracy of 85.19% when predicting driving intentions for drowsy subjects, which shows its promising potential for mitigating traffic accidents related to drowsy driving. Notably, our method maintains over 75% accuracy when more than half of the channels are missing or corrupted, underscoring its adaptability in real-life driving.

Figures (7)

• Figure 1: Illustration of the driving intention prediction task. The driver starts from straight proceeding on the road before the deviation occurs at a random time (deviation onset). After the deviation onset, the driver needs to take action to steer back to the original lane. Finally, the car is back in the original lane and the driver focuses on driving straight.
• Figure 2: Illustration of the model structure, the EEG waves $\mathbf{e}$ are first converted into frequency spectrogram $\mathbf{w}$ using Welch's method. Afterward, we patchify $\mathbf{w}$ and perform random masking to the EEG tokens using channel masking or frequency masking strategies. Subsequently, a transformer-based encoder processes the unmasked EEG tokens into the latent representation $\mathbf{z}$. Then, a decoder is used to reconstruct the spectrogram $\hat{\mathbf{w}}$ using the latent representations and the mask tokens. For intention prediction, a classifier is used to classify the latent representations into different steering directions.
• Figure 3: Independent Components with associated PSD plots from central-frontal and parietal areas.
• Figure 4: Ablation study on multitask learning and effect of our pretrained weights
• Figure 5: Confusion matrices of MEM model on different vigilance states. R: right turning intention, L: left turning intention, S: straight proceeding intention.