Driver Drowsiness Detection Based On ResNet-18 And Transfer Learning

To solve a common problem that the recognizing accuracy of vehicle drivers' tiredness system depends too much on a large number of training datasets, which are hard to acquire, we present a model that introduces the transfer learning theory and the Resnet-18 model. Firstly, after profoundly analyzing the drivers' head and facial expressions in a drowsy situation, we define three drivers' essential fatiguing characteristics and construct a system to collect them. Afterwards, we focus on setting up datasets and building a system to recognize and classify fatiguing characteristics based on the Resnet-18 transfer learning neural network model. Finally, the experiment's result verifies the effectiveness of proposed algorithms. It shows that the system's accuracy reaches 98.05%, which meets the standard of high accuracy.