Effective Fatigue Driving Detection by Machine Learning

Fatigue driving is a global road hazard. There are many possible causes for fatigue driving, but many drivers may not be aware of the situation. Therefore, it is important to detect fatigue driving and alert the driver to avoid potential damage. Quite a few techniques have been developed over the years for the detection of fatigue driving. In this paper, a machine learning approach is proposed for the purpose. The method combines several features used in determining the state of drivers. Machine learning usually involves large amounts of data, which makes it impractical for implementation on resource-restrained platforms. In this paper, we introduce ways to reduce the amount of training data needed. Then, two machine learning models are proposed, one with several fully connected layers, whereas the other replaces the layers with long short-term memory (LSTM). The lightweight structure simplifies the overall complexity and the number of parameters considerably. The two models are tested on the YawDD dataset. Results reveal that the two have comparable performance in fatigue driving detection, with an accuracy of close to 99% for both models.