A Context-Aware Driver Assistance Framework Using Multimodal Attentiveness Monitoring and TTC-Based Risk Fusion

Abstract

This study proposes a driver-monitoring and active-safety framework that integrates multimodal driver-state assessment with environmental risk evaluation. Driver attentiveness is estimated from visual and tactile cues, including eye behavior, head pose, hand posture, and steering-wheel touch interaction, while an external perception module detects frontal obstacles and estimates range using a vision-based deep learning approach. Based on the estimated range and relative closing velocity, Time-to-Collision (TTC) is calculated to represent collision urgency. A scoring-based fusion strategy combines the attentiveness score and TTC into a unified safety decision state that adjusts vehicle-control intervention according to both driver condition and environmental risk. The framework is implemented on a distributed embedded architecture and evaluated on a 1:10-scale vehicle platform under focused, distracted, and critical driving conditions. Experimental results show that incorporating TTC changes intervention behavior, with the largest descriptive difference observed under distracted driving and only a small additional effect under the critical condition. None of the paired Score-only versus Score–TTC comparisons reached statistical significance. Accordingly, the results are interpreted as proof-of-concept evidence that driver-state information and collision-risk information can be integrated within a unified closed-loop intervention framework, rather than as evidence of statistically established safety superiority. The proposed architecture provides a practical basis for further validation under more representative vehicle and traffic conditions.