Distracted driving recognition using Vision Transformer for human-machine co-driving

In order to achieve human-machine co-driving, the accurate and timely recognition of driving behavior is the first problem that needs to be solved. Numerous traffic accidents are caused by distracted driving behaviors, leading to the study of distracted driving recognition as an important topic in the traffic field. To overcome the shortcomings of existing researches, such as the low accuracy due to the insufficient data or the poor real-time performance due to lengthy layers of deep neural networks, we proposed a distracted driving recognition model based on the finetuned Vision Transformer, called DDR-ViT-finetuned. The model was trained and tested on the State Farm dataset compared to other methods. The experimental results demonstrated that the novel model achieved the highest accuracy rate of 97.5%.