Vision-Based Driver's Cognitive Load Classification Considering Eye Movement Using Machine Learning and Deep...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Road fatalities are increasing due to unsafe driver behaviors.
• Driver alertness is critical for traffic safety in both human and autonomous vehicles.
• Assessing cognitive load non-invasively is challenging, with wired sensors being impractical.

Purpose of the Study:

• To develop a non-contact, vision-based method for assessing driver cognitive load.
• To extract relevant features from driver eye movements.
• To classify driver cognitive load using machine learning and deep learning models.

Main Methods:

• Utilized image processing to analyze driver eye movement signals.
• Implemented manual feature extraction based on domain knowledge.
• Employed automatic feature extraction using deep learning architectures.
• Developed and compared five machine learning and three deep learning models.

Main Results:

• Achieved a maximum classification accuracy of 92% using a support vector machine with a linear kernel.
• Obtained 91% accuracy with a convolutional neural network model.
• Demonstrated the effectiveness of non-contact eye-tracking for cognitive load assessment.

Conclusions:

• Non-contact eye-tracking technology offers a promising solution for monitoring driver alertness.
• This method can be integrated into advanced driver-assistance systems (ADAS) to enhance road safety.
• The developed models provide accurate classification of driver cognitive load.