Online SVM-based personalizing method for the drowsiness detection of drivers.

Area of Science:

• Physiological signal processing
• Machine learning for driver monitoring
• Human-computer interaction in automotive safety

Background:

• Inter-driver variation poses a significant challenge for drowsiness detection systems relying on physiological signals.
• Existing systems often struggle with individual differences, leading to reduced accuracy and reliability.
• Personalization is crucial for effective drowsiness detection in real-world driving scenarios.

Purpose of the Study:

• To propose and validate an online support vector machine (OSVM)-based method for personalized drowsiness detection.
• To address the challenge of inter-driver variation in physiological signal-based drowsiness detection.
• To enhance driver safety through more accurate and adaptive drowsiness monitoring.

Main Methods:

• Development of an online support vector machine (OSVM) algorithm for adaptive personalization.
• Utilizing user-specific feedback data to retrain and refine the OSVM model.
• Implementation of a switching mechanism between two OSVMs for improved initial accuracy and adaptation speed.
• Validation using data from wearable devices and an indoor driving simulator.

Main Results:

• The proposed OSVM method significantly improved average detection accuracy from 72.05% to 95.66% across 28 subjects.
• Personalization through feedback data demonstrated effective adaptation to individual driver characteristics.
• The switching method contributed to lower initial errors and faster adaptation rates.

Conclusions:

• The developed OSVM-based method effectively personalizes drowsiness detection systems, overcoming inter-driver variation.
• The approach enhances drowsiness detection accuracy and promises increased driver safety.
• Real-time adaptation using feedback data is key to robust and reliable driver monitoring systems.