With the development of intelligent transportation systems, traffic sign detection has become a crucial component in autonomous driving. However, building high-performance traffic sign detection models typically requires a large amount of real image data and annotations, which can be challenging under resource constraints. To address this issue, this paper proposes a method leveraging synthetic images to train models. By applying data augmentation techniques, traffic signs are synthesized onto various backgrounds to improve detection accuracy. Initially, a preliminary traffic sign detection model is trained using synthetic images, which is then employed to automatically label real images. The labeled results are further used to retrain the model. Additionally, we test the effect of confidence thresholds on detection precision, analyzing the impact of both the absence of a threshold and different threshold values on performance. Experimental results demonstrate that, without the need for extensive real annotated data, combining synthetic images with an automatic labeling system effectively enhances the performance of traffic sign detection.