基于改进YOLOv4的交通标志识别方法

Traffic sign recognition method based on improved YOLOv4

文中提出了一种基于改进YOLOv4的交通标志识别方法,用于解决当前在交通标志识别任务中基于深度学习方法实时性较差、准确度较低的问题。在原版YOLOv4网络架构的基础上,使用原始的Darknet残差层替代了CSPDarknet53的CSP阶段,并对YOLOv4的PAN体系进行了CSP化,降低了运算量。用改进后的YOLOv4算法进行交通标志的特征提取,经过迁移学习对模型进行调整后实现了道路环境下交通标志的识别。为了测试改进算法的性能,在TT100K交通标志数据集上进行相关识别任务实验。实验结果表明,该算法的平均水平精度值(mAP)达到了86%,相较于原版YOLOv4算法提升了2.6%,每秒帧数(FPS)相较于原版YOLOv4算法提升了4.1。改进算法在检测精度和检测速度上较原版算法均有一定的提升。

Aiming at the problems of poor real⁃time performance and low accuracy in current deep learning⁃based traffic sign recognition tasks,an improved traffic sign recognition method based on YOLOv4 is proposed.On the basis of the original YOLOv4 network architecture,the original Darknet residual layer is used to replace the CSP stage of CSPDarknet53,and the PAN architecture of YOLOv4 is CSP-ized to reduce the amount of computation.The improved YOLOv4 algorithm is used to extract the features of traffic signs,and the model is adjusted through migration learning to realize the recognition of traffic signs in the road environment.In order to test the performance of the improved algorithm,the relevant recognition task experiment was carried out on the TT100K traffic sign data set.Experimental results show that the mean Average Precision(mAP)of the algorithm has reached 86%,which is an increase of 2.6 percentage points compared with the original YOLOv4 algorithm,and the number of Frames Per Second(FPS)has increased by 4.1 compared with the original YOLOv4 algorithm.The improved algorithm has a certain improvement in detection accuracy and detection speed compared with the original algorithm.