Abstract: Aiming at the problems of small target size and low detection accuracy in traffic sign detection, especially in the case of long-distance shooting and serious occlusion, traditional detection algorithms are often difficult to accurately identify traffic signs. In this paper, a traffic sign detection algorithm based on improved RT-DETR is proposed. First. Considering the scarcity of current datasets in the case of occluded traffic signs, this paper builds a self-constructed dataset of traffic signs under occluded conditions. Then, a lightweight composite inflated residual block is constructed to replace the BasicBlock in the original backbone extraction network by introducing an inflated reparameter block in the inverse residual shifted block, which enhances the feature extraction capability of the model. Finally, the loss function of the RT-DETR model is optimized, and the Inner-MPDIoU joint loss function is proposed to accelerate the convergence speed of the model. The experimental results show that the improved algorithm has a mAP of 94.2% on the homemade dataset, which is improved by 4.7% compared to the original algorithm, respectively, and the mAPs on the publicly available datasets, TT100K and CCTSDB2021, are 92.8% and 91.7%, which are improved by 3.1% and 2.4% compared to the original algorithm, and the Params and FLOPs are improved by 26.0% compared to the original algorithm by 26.0% and 12.5%, respectively. It is proved that the improved method proposed in this paper greatly reduces the amount of computation and the number of parameters, and effectively improves the detection accuracy of traffic signs under the occlusion situation