针对现有的目标检测模型对复杂天气下的交通标志检测存在漏检与错检的情况,提出了一种改进YOLOv5s的交通标志识别算法.为提高算法在各种复杂场景下的适应性,设计了一种基于重参数化(Re-Parameterized,ReP)的C3模块,将其命名为C3_DB;在网络的Neck部分引入上下文聚合模块,使得算法可以聚焦于重点区域的特征,减少复杂背景造成的混淆,从而提升模型的特征提取能力;引入高效交并比(Efficient Intersection over Union,EIoU)损失函数代替传统的完全交并比(Complete Intersection over Union,CIoU)损失函数,提升模型训练时的收敛速度,进一步提升模型对复杂情景下目标的检测性能.在中国交通标志数据集CCTSDB 2021上的实验结果表明,改进后算法的平均精度均值(mean Average Precision,mAP)为79.8％,相较于YOLOv5s提升2.4％,检测速度达到128帧/秒,在检测性能与检测速度之间取得了较好的平衡.意味着改进后的算法能够满足更高精度的交通标志实时检测需求,为实际应用提供了可靠的解决方案.研究成果对未来的交通管理和自动驾驶系统的发展具有重要意义,为提高安全性和可靠性提供了新的前景.
Traffic Sign Detection Algorithm Based on Improved YOLOv5s
To address the issue of missed and false detections of traffic signs under complex weather conditions by existing target detection models,an improved traffic sign recognition algorithm based on YOLOv5s is proposed.To enhance the adaptability of the algorithm in various complex scenarios,a Re-Parameterized(ReP)C3 module called C3_DB is designed.By introducing a context aggregation module in the Neck part of the network,the algorithm focuses on key region features,reducing confusion caused by complex backgrounds,thereby improving the model's feature extraction capability.The Efficient Intersection over Union(EIoU)loss function is introduced in replacement of the traditional Complete Intersection over Union(CIoU)to enhance the model's convergence speed during training and further improve the target detection performance under complex scenarios.Experimental results on Chinese Traffic Sign Dataset CCTSDB 2021 show that the improved algorithm achieves a mean Average Precision(mAP)of 79.8％,which is increased by 2.4％compared to YOLOv5s,with a detection speed of 128 frame/second.This balance between detection performance and speed suggests that the improved algorithm can meet the higher precision requirements for real-time detection of traffic signs,providing a reliable solution for practical applications.The research achievement is of great significance for the future development of traffic management and autonomous driving systems,providing new prospects for improvement of safety and reliability.
万方数据
维普
