VisionS: Real-Time Data-Driven Adaptive Driver Monitoring and Awareness System for Safer Roads.

Driver distraction and drowsiness are leading causes of road accidents worldwide. With new EU regulations mandating to have strong monitoring systems for enhancing road safety, traditional Driver Monitoring Systems (DMS) are typically marred by problems like false alarms, low flexibility, and weak on-the-fly performance. The current work focuses on designing VisionS, an adaptive on-the-fly driver monitoring system that improves distraction and drowsiness detection using computer vision and predictive analytics through data fusion with a physiological-based DMS, PredictS, realized through off-the-shelf Garmin wearable devices. Based on head pose dynamics in combination with behavioral patterns, VisionS provides precise on-the-fly feedback, producing timely alerts to offset unsafe driving behavior. Distraction and drowsiness detection is hard because there is variability in behavior, environment, and sensor capabilities. Vision-based methods can incorrectly label normal head movement as distraction, and drowsiness detection models have no redundancy and hence have uncertain classifications. This work bridges these gaps by proposing a hybrid system: VisionS continuously monitors head pose dynamics for distraction detection and employs statistical movement analysis to cross-validate drowsiness detection with PredictS. The primary contribution of this work is that the use of multiple sources of data improves accuracy and reliability of driver state estimation. VisionS uses computer vision techniques for real-time face tracking and head pose estimation. The system detects distraction by monitoring attention-related patterns of behavior and alerts visually or audibly when it detects signs of inattentiveness. To detect drowsiness, it monitors variability in movement and cross-verifies outcomes with PredictS to optimize and improve the sensitivity and specificity of the system. The system is implemented on an Android platform to enable real-time operation, providing immediate feedback on driver states. Validation proved that VisionS accurately identifies drowsiness and distraction via adaptive attention pattern and head movement monitoring and that real-time warning reduced reaction time to threats and that predictive modeling enhanced detection accuracy. The system outperformed baseline single-sensor methods consistently in discriminating against natural movement, inattention, and impairment from fatigue. This research confirms VisionS as an end-to-end, real-time driver monitoring system that accurately detects distraction and drowsiness. Behavioral analysis combined with predictive modeling achieves increased robustness and decreased misclassifications. Future research will investigate multi-sensor fusion and adaptive alerting to enhance performance and user experience further.