彭 博,臧 笛.基于深度学习的车标识别方法研究[J].计算机科学,2015,42(4):268-273
基于深度学习的车标识别方法研究
Vehicle Logo Recognition Based on Deep Learning
  
DOI:10.11896/j.issn.1002-137X.2015.04.055
中文关键词:  肇事车辆,车标,深度学习,车标识别
英文关键词:Accident-causing vehicle,Vehicle logo,Deep learning,Vehicle logo recognition
基金项目:本文受科技部国际合作专项(2012DFG11580),国家自然科学基金(61103071),教育部博士学科点新教师基金(20110072120065),留学回国人员科研启动基金资助
作者单位E-mail
彭 博 同济大学计算机科学与技术系 上海201804同济大学嵌入式系统与服务计算教育部重点实验室 上海200092  
臧 笛 同济大学计算机科学与技术系 上海201804同济大学嵌入式系统与服务计算教育部重点实验室 上海200092 zangdi@tongji.edu.cn 
摘要点击次数: 10227
全文下载次数: 2580
中文摘要:
      对交通监控录像中车牌污损、遮挡的肇事车辆信息进行确认是现阶段智能交通系统中的一个重要问题,车标作为一个关键特征,可以起到辅助判别的作用。 提出了一种基于深度学习的车标识别方法,相对于以人工提取特征为主的传统车标识别方法,该方法具有可自主学习特征、可直接输入图像等优点。实验表明,这种方法正确率较高,在光照变化和噪声污染下的准确性和稳定性较好,能够有效降低车标识别的错误率。
英文摘要:
      Identification of vehicles which have caused traffic accidents is an important issue in intelligent transportation system (ITS).However,due to the missing or the stains on the car license plates,it is difficult to locate the vehicles.Logos,as key features of vehicles,can also help to identify vehicles of interest.This paper proposed a method to detect and recognize vehicle logos based on deep learning.Compared with traditional approaches that extract features manually,this method uses original images as inputs to learn features automatically.Experimental results demonstrate that the proposed method has a high accuracy,and even in the conditions of illumination change and noise contamination,it is able to produce stable and accurate results.
查看全文  查看/发表评论  下载PDF阅读器