Study on Distracted Driving Behavior based on Transfer Learning

At present, many traffic accidents are caused by distracted behaviors of drivers while driving. For the detection of distracted behaviors of drivers, as the pre-training model has the ability to detect low-dimensional information of images, we adopt a vision-related transfer learning method. This paper compares the actual effect of improved training of driving distraction model based on VGG16 and ResNet50 pre-training model, and finally uses VGG16 network to build a driving distraction detection model. The structure of the full connection layer of the original model was changed, and the influence of different numbers of neurons in the full connection layer on the model performance was compared. Finally, the full connection layer of 128 neurons was taken as the best model to detect and classify 10 driving behaviors in the data set, and the accuracy of the model reached 98.5% in the test set. The results show that the trained model has the ability to preliminarily detect whether the driver has distracted behavior.