In view of a series of problems such as large error and slow detection speed of traffic signs under natural scenes in China.An improved YOLOV4 algorithm is proposed.Firstly, image enhancement, image denoising and other processing are added to the input end of the algorithm. Then, the detection layer of the algorithm is modified by deleting the 19*19 detection layer and adding the 152*152 detection layer.Finally, the K-meansⅡ clustering algorithm is used to carry out clustering analysis on the reconstructed traffic sign data set, and the initial candidate box of the network is redefined.The experimental results show that the improved algorithm can detect small traffic signs accurately and in real time in the natural scene.96mAP is obtained on the traffic sign data set based on the CCTSDB data set, and the detection speed is 26FPS, which is 1.7% and 1.4 higher than YOLOV4 algorithm respectively.