Driving Fatigue Identification Based on Multiple Non-intrusive Indicators

Fatigued driving accidents have a high fatality rate. It is necessary to identify the driver's fatigue state. Traditional classification methods such as support vector machine take a long time to solve high-dimensional data classification, leading to high time delay. In order to reduce the time delay while ensuring the accuracy of fatigue identification, this research compared the classification results of support vector machine and artificial neural network. A fatigued driving simulation experiment was designed to collect driver's personal information, eye movement, physiology, and driving performance data. Support vector machine and artificial neural network were adopted to identify the fatigue state and compare the identification accuracy. The result indicated that artificial neural network could identify the driver's fatigue state more quickly and perform a higher accuracy than support vector machine. Personal information and physiology data both play roles in improving the identification accuracy of the model. The data set containing personal information and physiological features can accurately and quickly identify the fatigue state of drivers with fewer indicators. The method has high accuracy, convenient collection, and low time delay. It can be widely used in the field of fatigue identification to prevent fatigued driving accidents.