Detecting slow eye movements using multi-scale one-dimensional convolutional neural network for driver sleepiness...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Biomedical Signal Processing

Background:

• Slow eye movements (SEMs) are critical indicators of driver sleep onset during simulated driving.
• Accurate detection of SEMs is essential for developing advanced driver sleepiness detection systems.

Purpose of the Study:

• To propose a novel Multi-Scale One-Dimensional Convolutional Neural Network (MS-1D-CNN) for classifying SEM waveforms.
• To evaluate the effectiveness of the MS-1D-CNN model in detecting driver sleep onset.

Main Methods:

• Developed an MS-1D-CNN model that utilizes multiple down-sampling branches and local convolutional layers to extract multi-scale features from SEM signals.
• Evaluated the model's performance using subject-subject and leave-one-subject-out cross-validation on standard train-test and continuous datasets.
• Compared the MS-1D-CNN performance against a baseline method using hand-designed features.

Main Results:

• Achieved high classification accuracies, around 99.1% and 98.6% on standard datasets, and 99.3% and 99.2% on continuous datasets, respectively.
• Demonstrated superior performance compared to the baseline method, with an average accuracy improvement of 3.5%.
• The MS-1D-CNN model showed robust performance even in leave-one-subject-out cross-validation scenarios.

Conclusions:

• The multi-scale feature extraction in the MS-1D-CNN model significantly enhances classification accuracy for SEM detection.
• The proposed MS-1D-CNN model offers a highly effective and reliable solution for detecting driver sleepiness.
• This technology has strong potential for real-world application in driver safety monitoring systems.