Enhancing Driver Stress Detection through Multimodal Integration of Eye Tracking and Physiological Signals

Widayat, Tri Agung; Mintje, Quirina Ariantji Patrisia; Yosepha, Sri Yanthy · 2025 · Crossref

DOI: 10.61978/logistica.v3i3.1147

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 critical safety issue of driver stress, which impairs attention, decision-making, and reaction time. Traditional unimodal monitoring methods often lack the sensitivity required for accurate real-time detection. To overcome these limitations, the authors propose and validate a novel multimodal framework that integrates synchronized eye-tracking and physiological data. The research aims to demonstrate that combining visual and autonomic markers yields higher classification accuracy than single-modality approaches, thereby providing a robust foundation for intelligent Driver Monitoring Systems (DMS). The experimental design involved thirty licensed drivers aged 21–45, recruited with balanced gender representation and no history of neurological or cardiovascular disorders. Participants underwent simulated driving tasks in a high-fidelity simulator under two conditions: a baseline relaxed state and a stress-induced state involving time pressure, simulated hazards, and cognitive load via secondary tasks. Data were collected using a Smart Eye Pro system (120 Hz) for eye-tracking metrics (pupil diameter, fixation duration, blink rate, gaze spread) and wearable sensors for physiological signals (heart rate, galvanic skin response, heart rate variability). All data streams were timestamped and synchronized. The researchers extracted key features and analyzed them using Linear Discriminant Analysis (LDA), benchmarked against Support Vector Machines and Random Forests, with 10-fold cross-validation to ensure robustness. The results revealed distinct physiological and visual changes under stress. Eye-tracking metrics showed a 20% increase in pupil diameter, a 35% rise in blink rate, and a significant narrowing of gaze spread, indicative of visual tunneling. Fixation duration decreased by 47%, likely due to rapid attention shifts required by the stressors. Physiologically, heart rate increased by 17%, skin conductance rose by 31%, and heart rate variability decreased by 19%, reflecting sympathetic nervous system arousal. In terms of classification performance, the unimodal eye-tracking model achieved 82.3% accuracy, while the physiological-only model reached 84.7%. The combined multimodal model significantly outperformed both, achieving 91.4% classification accuracy. The study concludes that multimodal integration enhances the sensitivity and reliability of driver stress detection by capturing complementary visual and autonomic indicators. These findings support the development of adaptive DMS capable of real-time stress recognition and proactive safety interventions. However, the authors note that the modest sample size limits generalizability and highlight challenges for real-world deployment, including environmental variability, sensor calibration, and ethical concerns regarding privacy and data governance. Future work should focus on large-scale on-road validation, personalized calibration strategies to account for individual variability, and the implementation of ethical frameworks to ensure responsible use of biometric monitoring technologies.

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
• workload measurement
• stress driving
• gaze based attention detection
• distraction detection algorithms
• dms validation

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, validation psychometrics