Driver Distraction Detection Based on Improved Swin Transformer

Distracted driving constitutes a primary factor leading to traffic accidents in contemporary society. To mitigate the occurrence of traffic accidents and enhance driving safety, an advanced SE-Swin Transformer network is proposed for the detection of driver distraction behaviors. Given the significant individual differences among distracted driving behaviors, this method integrates the Squeeze-and-Excitation (SE) module into the Swin Transformer structure, thereby augmenting the model’s adaptability and recognition accuracy for diverse driver behavior patterns. Utilizing the StateFarm dataset for training and validation, this model ensures its efficacy in real-world applications. Through an in-depth analysis of various distracted driving behaviors within the StateFarm dataset, the developed method is capable not only of identifying common types of distractions but also of adapting to specific behavioral habits of individual drivers. The experimental results show that the identification accuracy of the distracted driving detection method proposed in this study is 97.23%, which is significantly improved compared wtih the comparision methods. The findings further suggest that considering individual differences can enhance the method’s accuracy and generalizability.