Abstract: At present, there are a series of problems in the current task of Positioning and classification traffic signs in high-definition street view images. The image target background is complicated and the small targets are difficult to recognize. A new traffic sign recognition algorithm based on YOLOV4 model is proposed. Firstly, a new type of attention mechanism CA module is embedded in the original backbone network to form a pair of direction-aware and position-sensitive attention maps, so that the network can focus on effective features in a larger area. Secondly, features in the neck are enhanced Add an appropriate amount of RFB modules to the network to further enhance the feature integration capability of the network. Experiments on the TT100K traffic sign data set found that when the IOU threshold of the improved algorithm is 0.5, the mAP index of the improved algorithm increased by nearly 4%, and the number of detected frames per second reached 39FPS. Therefore, the CA-RFB-YOLOV4 algorithm proposed in this paper can significantly improve the recognition accuracy of target data with a small increase in calculation parameters, and better meet the needs of traffic sign recognition in actual scenes.