Fast detection algorithm based on YOLOv8 with structural reparameterisation and grouped convolutional heads

As autonomous driving technology advances swiftly, the precise identification of road signs has become essential for enabling autonomous driving. This paper introduces an enhanced YOLOv8 model for road sign recognition. This model enhances the original C2f module by incorporating the RepViT Block, resulting in the C2f_RVB module. This alteration not only hastens the convergence rate of the model but also markedly boosts detection precision. Furthermore, we introduce a highly efficient architectural framework utilizing lightweight ViTs, which augments the model's detection speed while minimizing the parameter count. Lastly, we incorporate grouped convolution to enhance the detection head of YOLOv8. Tests were conducted utilizing the Street Sign Locator dataset, which was improved with information from the Roboflow stage. The exploratory comes about illustrate that the proposed RVB-YOLOv8 demonstrate accomplishes critical upgrades in discovery measurements like mAP and FLOPs when compared to the initial YOLOv8 demonstrate, whereas too diminishing the number of parameters by one-third.