Lightweight detection method of distracted driving based on YOLO-SSC

To address the problems of high device requirements, low detection accuracy, and high miss detection rate of overlapping targets in traditional distracted driving target detection models, a vehicle target detection algorithm called YOLO-SSC, based on the improved YOLOv8, was proposed. First, the main body network was reconstructed by using multi-scale fusion. Then, the StarBlock module was used to improve the C2f module in the neck, achieving multi-scale feature fusion to improve the model's detection accuracy for different size and distance targets. Finally, a lightweight model with shared convolution was used to further reduce the model's parameter number and computational cost. The experimental results show that the mAP50 of the improved model is 96.3%, the number of model parameters and the amount of model calculation are 1.4M and 4.5G, respectively, which are reduced by 53.3%and 44.4%compared with YOLOv8n.