페이스 랜드마크 트래킹과 메타휴먼 재구성을 활용한 AI 기반 운전자 졸음 감지

초록

Drowsy driving poses a significant risk, resulting in an average of 2.9 fatalities per 100 accidents—nearly twice the 1.5 fatalities associated with drunk driving. Cognitive function declines when in-vehicle CO2 levels exceed 2,000 ppm, indicating the necessity for real-time drowsiness detection. This study proposes an AI-based face-tracking system that integrates drowsiness detection with CO2 measurement. Infrared cameras within VR HMDs capture subtle facial muscle movements, which are transmitted to a metahuman model within a 3D engine via Live Link. A supervised model, trained on 2,400 metahuman images, underpins drowsiness detection. The metahuman background color dynamically adjusts in response to CO2 concentration, facilitating intuitive monitoring. If prolonged drowsiness is detected, the system issues a warning. The performance of the AI model was validated using k-fold cross-validation and mean average precision. This approach enables real-time driver monitoring by delivering multistage warnings, immediate feedback, and vehicle control interventions when necessary.

키워드