Highway Vehicle Object Detection System Based on RepViT and YOLOv8

With the development of expressway, the traffic flow increases sharply, which is of great significance to the statistics of the number and types of vehicles on the road. This paper proposes an efficient design architecture based on YOLOv8 and integrated Lightweight Vision Transformer (RepViT). Current studies have shown that ViTs show better performance and lower latency than lightweight convolutional neural networks CNNs on resource-constrained mobile devices. Through the efficient architecture design integrating lightweight ViTs, the standard lightweight CNNs are gradually enhanced, and a new pure lightweight CNNs, namely RepViT, is obtained. By using RepViT instead of YOLOv8’s original backbone network, the model can obtain higher accuracy and training speed.