Driver Distraction Detection Algorithm Based on High-Order Global Interaction Features

Distracted driving behavior is one of the main causes of road traffic safety problems. In view of the problems of high model complexity, unstable detection performance and high hardware cost of existing distraction detection algorithms, this paper proposes a driver distraction detection algorithm based on high-order global interaction features. First, the backbone network is reconstructed using the C3-HB module, in which the HorNet recursive gated convolution is used to learn the long-range dependencies of the image and obtain high-order features, and the bottleneck in the C3 module is replaced by Hornet Block, which improves the detection accuracy of small targets and complex scenes; secondly, the global parallel attention mechanism PGAM is designed to enhance the perception ability of global interaction features and reduce local information loss; finally, the loss function of YOLOv5s is replaced by α-CIOU, which effectively balances the difficult and easy samples and further optimizes the detection effect of driver distraction. Experiments show that the model proposed in this paper achieves an mAP50 of 97.32% on the State Farm dataset and the self-built dataset. While improving the detection accuracy, the number of parameters is only 8.68M and the computational complexity is 12.80GFLOPs, showing excellent comprehensive performance and is suitable for tasks that require real-time and high precision.