Dual Modal Physiological Fusion for Five-State Driver Impairment Detection

Abstract

Current driver monitoring systems detect drowsiness and distraction but not anger, even though aggressive driving accounts for 56% of fatal crashes (AAA Foundation for Traffic Safety, 2016). This paper presents a five-state driver impairment classification system using late fusion of ECG-derived heart rate variability and facial action unit (AU) analysis. Three contributions are reported. First, an operational pipeline using BIOPAC MP150 ECG and OpenFace 2.0, integrated with a fixed base driving simulator, is fully implemented. Second, Monte Carlo simulation ( N  = 200 agents; 500 replications; pilot n  = 24) projects that Bayesian late fusion achieves F 1 = 0.85 for anger detection, a +0.13 gain over the best unimodal system, with graceful degradation under noise. Third, the Human Identity and Autonomy Gap (HIAG) framework predicts that a planned companion transparency study ( N  = 120) will find that anger monitoring will elicit greater psychological reactance than monitoring of functional states.