Abstract: Object detection, a pivotal discipline within computer vision, has seen extensive deployment across various sectors, notably in the precise parsing of vehicular traffic contexts, where its critical function is paramount. However, in road traffic environments with dense vehicles and complex road conditions, traditional object detection methods have problems such as insufficient accuracy, misjudgment, and omission. In addition, the deployment challenges of large-scale models on resource constrained edge devices further limit their application in practical scenarios. A method for road object detection based on the YOLOv8 object detection model is proposed. Firstly, the GhostConv convolution kernel is used to replace the conventional convolution of the Bottleneck layer in the C2f structure, significantly reducing the parameter size of the model. Then, by integrating GhostConv and LSK attention mechanism, a GhostLSK module is constructed, which significantly improves the model's feature extraction ability for traffic vehicles and pedestrians while reducing the overall parameter size. Finally, add a small object detection layer on the detection head to enhance the detection capability of small objects.