正在加载图片...
第10卷第5期 智能系统学报 Vol.10 No.5 2015年10月 CAAI Transactions on Intelligent Systems 0ct.2015 D0I:10.11992/is.201406044 网s络出版地址:htp://ww.cmki.net/kcms/detail/23.1538.tp.20150930.1556.010.html 半监督SVM分类算法的交通视频车辆检测方法 蒋新华12,高晟3,廖律超12,邹复民2 (1.中南大学信息科学与工程学院,湖南长沙410075:2.福建工程学院福建省汽车电子与电驱动技术重点实验 室,福建福州350108:3.中南大学软件学院,湖南长沙410075) 摘要:针对交通场景运动车辆检测中车辆数目统计准确率不高、自适应性不强等问题,提出了一种基于半监督支 持向量机(SVM)分类算法的交通视频车辆检测方法。利用人工标记的少量样本,分别训练2个基于方向梯度直方 图(HOG)特征与基于局部二值模式(LBP)特征的不同核函数的SVM分类器:结合半监督算法的思想,构建SVM的 半监督分类方法(SEMI-SVM),标记未知样本并加人到原样本库中,该方法支持样本库动态更新,避免了繁重的人工 标记样本的工作,提高了自适应性:最后,通过三帧差分法提取运动区域,加载分类器在该区域进行多尺度检测,标 记检测出来的运动车辆,统计车辆数目。实验结果表明:该方法在具有一定的自适应性的同时,有较高的车辆检测 准确率,即使在复杂交通情况下,对运动车辆依然有很好的检测效果。 关键词:车辆检测:HOG特征:LBP特征:SVM分类器:半监督学习:运动区域 中图分类号:TP181文献标志码:A文章编号:1673-4785(2015)05-0690-09 中文引用格式:蒋新华,高晟,廖律超,等.半监督SVM分类算法的交通视频车辆检测方法[J].智能系统学报,2015,10(5):690- 698. 英文引用格式:JIANG Xinhua,,GAO Sheng,LIAO Lyuchao,etal.Traffic video vehicle detection based on semi-supervised SVM classification algorithm[J].CAAI Transactions on Intelligent Systems,2015,10(5):690-698. Traffic video vehicle detection based on semi-supervised SVM classification algorithm JIANG Xinhua'2,GAO Sheng,LIAO Ljuchao'2,ZOU Fumin2 (1.School of Information Science and Engineering,Central South University,Changsha 410075,China;2.Fujian Key Laboratory for Automotive Electronics and Electric Drive,Fujian University of Technology,Fuzhou 350108,China;3.School of Software Engineer- ing,Central South University,Changsha 410075,China) Abstract:This paper presents a kind of traffic video vehicle detection method based on a semi-supervised support vector machine (SVM)classification algorithm to improve accuracy and enhance adaptability of vehicle counting in the traffic scene.By analyzing a small number of artificially labeled samples,two SVM classifiers with different ker- nels are trained on the basis of histograms of oriented gradients (HOG)features and local binary pattern(LBP) features,respectively.A semi-supervised SVM(SEMI-SVM)for classification is proposed by adopting the thoughts of semi learning.Then the unknown samples are labeled and added into the original sample database.The proposed method supports data update of the dynamic sample database,avoids heavy manual work labeling samples and en- hances adaptability of the algorithm.A motion region is extracted using the three-frame difference rule.The classifi- er is then loaded to make a multi-scale detection in the extracted motion region,and moving vehicles are marked and counted.The results show the algorithm has good response,good adaptability,and the detection accuracy of moving vehicles is much improved,even under the complex traffic circumstances. Keywords:vehicle detection;histograms of oriented gradients (HOG)feature;local binary pattern LBP)fea- ture;support vector machine (SVM)classifier;semi-supervised learning;motion region 交通视频车辆检测是一种利用视频图像实现对 车辆进行检测的交通检测技术,它可以检测多种参 数和检测范围较大等优点,但如何设计高效的车辆 收稿日期:2014-06-22.网络出版日期:2015-09-30. 基金项目:国家自然科学基金资助项目(61304199,41471333):福 检测算法,提高检测准确率和实时性是亟待解决的 建省自然科学基金(201301214):福建省科技重大专项专 问题。 题资助项目(2011HZ0002-1):福建省交通科技计划项目 目前,通过交通视频进行车辆检测的方法主要 (201318):福建省教有厅B类科研项目(UB3213). 通信作者:高最.E-mail:csugaosheng@163.com. 有:帧间差分法山、灰度等级法[)]、背景相减法[3]第 10 卷第 5 期 智 能 系 统 学 报 Vol.10 №.5 2015 年 10 月 CAAI Transactions on Intelligent Systems Oct. 2015 DOI:10.11992 / tis.201406044 网络出版地址:http: / / www.cnki.net / kcms/ detail / 23.1538.tp.20150930.1556.010.html 半监督 SVM 分类算法的交通视频车辆检测方法 蒋新华1,2 ,高晟3 ,廖律超1,2 ,邹复民2 (1. 中南大学 信息科学与工程学院,湖南 长沙 410075; 2. 福建工程学院 福建省汽车电子与电驱动技术重点实验 室,福建 福州 350108; 3. 中南大学 软件学院,湖南 长沙 410075) 摘 要:针对交通场景运动车辆检测中车辆数目统计准确率不高、自适应性不强等问题,提出了一种基于半监督支 持向量机(SVM)分类算法的交通视频车辆检测方法。 利用人工标记的少量样本,分别训练 2 个基于方向梯度直方 图(HOG)特征与基于局部二值模式(LBP)特征的不同核函数的 SVM 分类器;结合半监督算法的思想,构建 SVM 的 半监督分类方法(SEMI⁃SVM),标记未知样本并加入到原样本库中,该方法支持样本库动态更新,避免了繁重的人工 标记样本的工作,提高了自适应性;最后,通过三帧差分法提取运动区域,加载分类器在该区域进行多尺度检测,标 记检测出来的运动车辆,统计车辆数目。 实验结果表明:该方法在具有一定的自适应性的同时,有较高的车辆检测 准确率,即使在复杂交通情况下,对运动车辆依然有很好的检测效果。 关键词:车辆检测;HOG 特征;LBP 特征;SVM 分类器;半监督学习;运动区域 中图分类号:TP181 文献标志码:A 文章编号:1673⁃4785(2015)05⁃0690⁃09 中文引用格式:蒋新华,高晟,廖律超,等. 半监督 SVM 分类算法的交通视频车辆检测方法[ J]. 智能系统学报, 2015, 10(5): 690⁃ 698. 英文引用格式:JIANG Xinhua, GAO Sheng, LIAO Lyuchao, et al. Traffic video vehicle detection based on semi⁃supervised SVM classification algorithm[J]. CAAI Transactions on Intelligent Systems, 2015, 10(5): 690⁃698. Traffic video vehicle detection based on semi⁃supervised SVM classification algorithm JIANG Xinhua 1,2 , GAO Sheng 3 , LIAO Ljuchao 1,2 , ZOU Fumin 2 (1. School of Information Science and Engineering, Central South University, Changsha 410075, China; 2. Fujian Key Laboratory for Automotive Electronics and Electric Drive, Fujian University of Technology, Fuzhou 350108, China; 3. School of Software Engineer⁃ ing, Central South University, Changsha 410075, China) Abstract:This paper presents a kind of traffic video vehicle detection method based on a semi⁃supervised support vector machine (SVM) classification algorithm to improve accuracy and enhance adaptability of vehicle counting in the traffic scene. By analyzing a small number of artificially labeled samples, two SVM classifiers with different ker⁃ nels are trained on the basis of histograms of oriented gradients (HOG) features and local binary pattern (LBP) features, respectively. A semi⁃supervised SVM (SEMI⁃SVM) for classification is proposed by adopting the thoughts of semi learning. Then the unknown samples are labeled and added into the original sample database. The proposed method supports data update of the dynamic sample database, avoids heavy manual work labeling samples and en⁃ hances adaptability of the algorithm. A motion region is extracted using the three⁃frame difference rule. The classifi⁃ er is then loaded to make a multi⁃scale detection in the extracted motion region, and moving vehicles are marked and counted. The results show the algorithm has good response, good adaptability, and the detection accuracy of moving vehicles is much improved, even under the complex traffic circumstances. Keywords:vehicle detection; histograms of oriented gradients (HOG) feature; local binary pattern ( LBP) fea⁃ ture; support vector machine (SVM) classifier; semi⁃supervised learning; motion region 收稿日期:2014⁃06⁃22. 网络出版日期:2015⁃09⁃3 基金项目:国家自然科学基金资助项目( 613041 通信作者:高晟. E⁃mail:csugaosheng@ 163.com 9 0 9 . ,41471333);福 . 交通视频车辆检测是一种利用视频图像实现对 车辆进行检测的交通检测技术,它可以检测多种参 数和检测范围较大等优点,但如何设计高效的车辆 检测算法,提高检测准确率和实时性是亟待解决的 问题。 目前,通过交通视频进行车辆检测的方法主要 有:帧间差分法[1] 、灰度等级法[2] 、背景相减法[3⁃5] 建省自然科学基金(2013J01214);福建省科技重大专项专 题资助项目(2011HZ0002-1);福建省交通科技计划项目 (201318);福建省教育厅B类科研项目(JB3213)
向下翻页>>
©2008-现在 cucdc.com 高等教育资讯网 版权所有