基于改进YOLOv5的复杂场景下交通标志识别方法
Traffic Sign Recognition Method in Complex Scenes Based on Improved YOLOv5
胡瑛 1刘狄昆 1刘拯 1李铸成 1乔汇东1
作者信息
- 1. 湖南工程学院 信息科学与工程学院,湘潭 411104
- 折叠
摘要
交通标志识别是智能驾驶不可缺少的重要环节,关系着人们进行智能驾驶的安全问题,本文以复杂环境下的交通标志为研究对象,针对目前交通标志识别难以兼顾实时性和准确率的问题,提出一种改进的YOLOv5交通标志识别算法.首先,对数据集做预处理与数据增强,加强对目标的检测能力;其次,引用PP-LCNet轻量型网络,减少主干网络参数量,实现模型轻量化;最后,在颈部网络融合注意力机制,以增强特征提取能力.实验结果表明,相较于原YOLOv5s模型,本文算法的模型参数量减少了25.9％,检测速度提高了50.08帧/s,平均精度达到97.58％,易于部署且能达到智能驾驶场景中对交通标志识别的实时性和准确率要求.
Abstract
Nowadays traffic signs recognition is an inevitable and important part of intelligent driving,which is related to the safety of people using intelligent driving.Therefore,this paper takes the traffic signs in complex environment as the research object,aiming at the problem that the current traffic signs recognition is difficult to take into account the real-time and accuracy,and proposes an improved YOLOv5 traffic signs recognition algorithm.First,preprocess and enhance the data set to strengthen the ability of target detection.Then,the PP-LCNet lightweight network is used to reduce the parameters of the backbone network,realize the lightweight of the model,integrate the attention mechanism in the neck network,and enhance the ability of feature extraction.Experiments show that compared with the original YOLOv5s model,the model parame-ters of this algorithm are reduced by 25.9％;the detection speed is increased by 50.08 frames/s,and the aver-age accuracy is 97.58％.It is easy to deploy and can meet the real-time and accuracy requirements of traffic signs recognition in intelligent driving scenes.
关键词
交通标志识别/数据增强/YOLOv5/PP-LCNet网络/注意力机制Key words
traffic signs recognition/data enhancement/YOLOv5/PP-LCNet/attention mechanism引用本文复制引用
基金项目
国家级大学生创新创业训练计划项目(202211342017)
湖南省教育厅优秀青年科研项目(22B0735)
湖南省自科基金面上项目(2022JJ30197)
湖南省教育厅重点科研项目(20A112)
出版年
2024
NETL
NSTL
万方数据