摘要
交通信号灯的在线识别是无人驾驶和辅助决策系统中的重要研究内容,文章给出了一种基于深度学习的交通信号灯识别和分类方法,该方法使用YOLO(You Only Look Once)模型,基于Microsoft COCO数据集的预训练模型进行二次迁移学习:先使用Bosch数据集进行迁移学习,再使用自制数据集做迁移学习。测试表明该方法训练后的模型具有较高准确率和实时性。同时,文章给出了基于检测结果提取综合路况信息的策略。
The online recognition of traffic lights is an important topic in unmanned and assisted decision-making systems.This paper presents a deep learning approach to traffic lights identification and classification.This method uses YOLO(You Only Look Once)model.The pre-training model of the COCO dataset is used for double-transfer learning:Bosch dataset is used for the first transfer learning,and self-made dataset is used for the secondary transfer learning.The test result shows that the model trained by this method has higher accuracy and real-time performance.Also,we presents a strategy for extracting comprehensive road condition information based on detection results.
作者
王莹
丁鹏
Wang Ying;Ding Peng(Wuxi Vocational and Technical College,Jiangsu Wuxi 214121)
出处
《汽车实用技术》
2018年第17期89-91,共3页
Automobile Applied Technology
基金
无锡职业技术学院校级课题(名称:无人车对交通信号装置的识别与道路选择研究
编号:3116013931)资助