Generative AI-Enhanced Multi-Physiological Signal Analysis for Intelligent Transportation Safety: An Attention-Masked Transformer Approach

The increasing prevalence of fatigue-related accidents in transportation systems necessitates advanced driver monitoring solutions to process complex physiological signals in real-time effectively. This paper presents generative AI-enhanced multi-physiological signal analysis (GEMPSA), a novel framework for intelligent transportation safety that leverages generative AI and transformer architectures to process multiple physiological signals for robust fatigue detection. Unlike existing approaches that process signals independently, GEMPSA introduces three key innovations: an integrative feature token for unified physiological state representation, a position-temporal encoding scheme that preserves spatial-temporal relationships, and a graph-based feature fusion approach that combines multiple physiological signals. This design enables end-to-end learning without extensive preprocessing while maintaining high accuracy in real-world conditions. Experiments across diverse driving conditions demonstrate GEMPSA’s superior performance, achieving 95.7% accuracy in fatigue detection, significantly outperforming existing methods. These results were obtained from extensive experiments using a dataset of 3,450 samples collected from 23 participants under controlled driving conditions, incorporating multi-modal physiological signals, including EEG, ECG, blood oxygen, and respiratory measurements. Statistical analyses through Nemenyi tests validate the framework’s consistent superiority across multiple metrics, while temporal and correlation analyses confirm its robust performance in real-world scenarios. The framework’s ability to maintain strong personal identification correlations further demonstrates its effectiveness in handling individual physiological variations, making it particularly suitable for practical deployment in intelligent transportation systems.