Aiming at the problems of high computational complexity and limited detection accuracy of existing distracted driving detection models, a lightweight distracted driving detection algorithm based on YOLOv8n was proposed. Firstly, the RevCol_faster network was used to reconstruct the YOLOv8n backbone, and the feature extraction ability was enhanced through the reversible column structure, while reducing the number of parameters and calculation. Secondly, the C2f_Star module was combined to optimize the neck structure and improve the multi-scale feature fusion ability to improve the detection accuracy of targets of different sizes and different distances. In addition, LSCD(lightweight shared convolution detection) was used to share the convolution detection head to reduce the calculation redundancy and improve the detection speed. Finally, the EIoU loss function was used to replace the original CIoU to enhance the shape perception of the target box and improve the positioning accuracy. Experimental results show that the improved model achieves 96.8% mAP50 on the Kaggle driver behavior dataset, reduces the number of parameters to 1.3 × 10⁶, reduces the amount of calculation to 3.0 × 10⁹, and reduces 57.60% and 56.67% respectively compared with YOLOv8n, and improves the inference speed by 4.5%. This study provides an efficient and computationally inexpensive solution for distracted driving detection in resource-constrained environments.