Research on Traffic Sign Detection Algorithm Based on YOLO

In view of the large amount of network parameters and calculations of the current two-stage target detection algorithm, and the low recognition accuracy of the single-stage target detection algorithm, an improved traffic sign detection algorithm based on yolov4-tiny is proposed.The algorithm uses the SSD structure idea to divide the ordinary convolution into two steps, which not only reduces the computing resources and the number of parameters, but also adds a new feature extraction structure to obtain more target features.Then, bottom-up multi-scale fusion is used, combined with low-level information to enrich the feature level of the network to improve feature utilization.Finally, CIoU is used as the bounding box regression loss function to speed up model convergence and improve the accuracy of traffic sign detection.The experimental results on the CCTSDB data set show that, compared with the original YOLOv4-tiny, the mAP is increased by about 12%, and the recognition speed is 10 ms. It indicates that the method proposed in this paper has a faster speed and higher accuracy.