Traffic sign recognition under complex road conditions is an important research content in the field of automatic driving, for the current traffic sign recognition of omission, false detection, and poor generalization under complex road conditions. This chapter proposes a traffic sign recognition algorithm based on improved YOLOv5. Firstly, the inverse perspective transformation is performed on the image captured by the onboard camera to convert the image from the fish-eye view to the normal perspective view to make the feature extraction elements more complete. To deal with multi-scale targets, a feature pyramid structure is introduced in YOLOv5. By adding additional convolutional layers to the top layer of the feature extractor, multiple feature maps with different resolutions can be obtained, representing the target information to varying scales of the image, so that the model can effectively recognize multi-scale targets. The experimental results show that compared with the original YOLOv5 algorithm, the improved algorithm improves the recognition speed and accuracy, is easy to implement, and can meet the accuracy and real-time requirements of traffic sign recognition under complex road conditions.