Case Study: Graz University of Technology detects driver drowsiness earlier with MathWorks AI tools

AI Knows When Drivers Are Getting Very Sleepy

Graz University of Technology, a research institution, faced the challenge of detecting driver drowsiness, a major cause of accidents, more accurately than existing methods. Their goal was to create a robust system that could classify different levels of drowsiness, which is crucial for the safety of Level 3 autonomous vehicles that require an alert driver to take over. To accomplish this, researchers at the university turned to MathWorks, utilizing MATLAB with its Wavelet Toolbox and Deep Learning Toolbox.

The solution implemented by MathWorks involved using deep learning to analyze electrocardiogram (ECG) signals. The team transformed the signals into wavelet scalogram images using MATLAB, which were then processed by a custom-built convolutional neural network to classify drowsiness into three states: alert, moderately drowsy, and extremely drowsy. The neural network, optimized with a MathWorks tool, achieved a 77-79% accuracy rate, significantly outperforming traditional methods that only reached 62-64% accuracy. This demonstrated to MathWorks and the automotive industry a substantially better way to perform driver drowsiness classification.

Graz University of Technology

Arno Eichberger

Engineering Professor