基于卷积神经网络的运动车辆视频检测方法
详细信息 查看官网全文
- 英文论文题名：The Method of Moving Vehicle Video Detection Based on Convolution Neural Network
- 论文作者：陈林凯 ; 阮雅端 ; 朱洪锦 ; 李昀峰 ; 陈启美
- 英文论文作者：Chen Linkai ; Ruan Yaduan ; Zhu Hongjing ; Li Yunfeng ; Chen Qimei ; School of Electronic Science and Engineering ; NanJing University ; College of Computer and Engineering ; Jiangsu University of Technology
- 年：2016
- 作者机构：南京大学电子科学与工程学院;江苏理工学院计算机工程学院;
- 论文关键词：车辆视频检测 ; 高斯混合模型 ; 卷积神经网络 ; 深度学习
- 英文论文关键词：vehicle detection ; Gaussian Mixture Model ; convolution neural network ; deep learning
- 会议召开时间：2016-06-24
- 会议录名称：2016年全国通信软件学术会议程序册与交流文集
- 语种：中文
- 分类号：TP391.41;TP183
- 学会代码：ZGTH
- 会议名称：2016年全国通信软件学术会议
- 会议地点：中国陕西西安
- 主办单位：中国通信学会
- 学会名称：中国通信学会
- 页数：6
- 文件大小：1709k
- 原文格式：O
- 会议级别：全国
摘要
运动车辆的检测现已采用视频检测技术,然而传统的高斯混合模型检测算法对背景环境过于敏感,光照变化、树叶的扰动都可能误检为运动车辆。为此,文中在介绍高斯混合模型建模、卷积神经网络的机理基础上,提出了一种高斯混合模型与卷积神经网络(CNN)结合的运动车辆检测方法:采用高斯混合模型对运动目标进行检测,通过车辆样本空间训练卷积神经网络模型,再用卷积神经网络对所检测出的运动区域目标特征提取,以区分出车或非车。经实际道路视频测试表明:与常用的稀疏编码(SC)和尺度不变特征转换(SIFT)特征提取方法相比,该算法的运动车辆检测的准确率分别提高了3.6%和7.9%。
Moving vehicle detection use video technology now,however,the traditional Gaussian Mixture Model(GMM) is sensitive to background environment,illumination changes and leaves disturbance may be lead to error detection.For this,the paper introduces the mechanism of the GMM and convolution neural network(CNN),and then purposed a vehicle detection method of GMM combined with CNN.Using GMM to detecting the moving object,training the CNN model through the vehicle sample spaces,then using CNN to extract feature from the detected moving object and classify to vehicle or nonvehicle.The experiment show that:this method can efficiently improve the accuracy of vehicle detection 3.6%and 6.9%,compared to common sparse code(SC) and SIFT methods,respectively.
Moving vehicle detection use video technology now,however,the traditional Gaussian Mixture Model(GMM) is sensitive to background environment,illumination changes and leaves disturbance may be lead to error detection.For this,the paper introduces the mechanism of the GMM and convolution neural network(CNN),and then purposed a vehicle detection method of GMM combined with CNN.Using GMM to detecting the moving object,training the CNN model through the vehicle sample spaces,then using CNN to extract feature from the detected moving object and classify to vehicle or nonvehicle.The experiment show that:this method can efficiently improve the accuracy of vehicle detection 3.6%and 6.9%,compared to common sparse code(SC) and SIFT methods,respectively.
引文
[1]刘操,郑宏,黎曦,等.基于多通道融合HOG特征的全天候运动车辆检测方法[J].武汉大学学报情息科学版,2015,40(8):1048-1053.
[2]Castrillon F.Guin A.Gucnslcr R Comparison of Modeling Approaches for Imputation of Video Detection Data in Intelligent Transportation Systems[J].Transportation Research Record:Journal of the Transportation Research Board.2012,2308(1):138-147.
[3]林立原,陈林.基于高斯混合模型和卡尔曼滤波的车辆检测与跟踪方法[J].武汉科技大学学报,2015(3):226-230.
[4]Chen L,Zhu P,Zhu G.Moving objects detection based on background subtraction combined with consecutive frames subtraction[C]//Future Information Technology and Management Engineering(FITME),2010International Conference on.IEEE,2010,2:545-548.
[5]Krizhevsky A,Sutskever I,Hinton G E.Imagenet classification with deep convolutional neural networks[C]//Advances in neural information processing systems.2012:1097-1105.
[6]Gupte S,Masoud O,Martin R F K,et al.Detection and classification of vehicles[J].Intelligent Transportation Systems,IEEE Transactions on,2002,3(1):37-47.
[7]R-CNN:Girshick R,Donahue J,Darrell T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C],CVPR,2014.
[8]Castrillon F,Guin A,Guensler R.Comparison of Modeling Approaches for Imputation of Video Detection Data in Intelligent Transportation Systems[J].Transportation Research Record:Journal of the Transportation Research Board,2012,2308(1):138-147.
[9]卢清华,吴志伟,范彦斌,等.基于混合高斯模型的运动车辆检测方法[J].光电子:激光,2013(4):751-757.
[10]Wang H,Cai Y,Chen L.A vehicle detection algorithm based on deep belief network[J].The scientific world journal,2014,2014
[11]Jia Y,Shelhamer E,Donahue J,et al.Caffe:Convolutional Architecture for Fast Feature Embedding[J].Eprint Arxiv,2014:675-678.
[12]Xiangjun C,Yaduan R,Peng Z,et al.Vehicle representation and classification of surveillance video based on sparse learning[J].Communications,China,2014,11(13):135-141.
[2]Castrillon F.Guin A.Gucnslcr R Comparison of Modeling Approaches for Imputation of Video Detection Data in Intelligent Transportation Systems[J].Transportation Research Record:Journal of the Transportation Research Board.2012,2308(1):138-147.
[3]林立原,陈林.基于高斯混合模型和卡尔曼滤波的车辆检测与跟踪方法[J].武汉科技大学学报,2015(3):226-230.
[4]Chen L,Zhu P,Zhu G.Moving objects detection based on background subtraction combined with consecutive frames subtraction[C]//Future Information Technology and Management Engineering(FITME),2010International Conference on.IEEE,2010,2:545-548.
[5]Krizhevsky A,Sutskever I,Hinton G E.Imagenet classification with deep convolutional neural networks[C]//Advances in neural information processing systems.2012:1097-1105.
[6]Gupte S,Masoud O,Martin R F K,et al.Detection and classification of vehicles[J].Intelligent Transportation Systems,IEEE Transactions on,2002,3(1):37-47.
[7]R-CNN:Girshick R,Donahue J,Darrell T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C],CVPR,2014.
[8]Castrillon F,Guin A,Guensler R.Comparison of Modeling Approaches for Imputation of Video Detection Data in Intelligent Transportation Systems[J].Transportation Research Record:Journal of the Transportation Research Board,2012,2308(1):138-147.
[9]卢清华,吴志伟,范彦斌,等.基于混合高斯模型的运动车辆检测方法[J].光电子:激光,2013(4):751-757.
[10]Wang H,Cai Y,Chen L.A vehicle detection algorithm based on deep belief network[J].The scientific world journal,2014,2014
[11]Jia Y,Shelhamer E,Donahue J,et al.Caffe:Convolutional Architecture for Fast Feature Embedding[J].Eprint Arxiv,2014:675-678.
[12]Xiangjun C,Yaduan R,Peng Z,et al.Vehicle representation and classification of surveillance video based on sparse learning[J].Communications,China,2014,11(13):135-141.