基于双路式卷积神经网络的车辆与行人检测
|
摘要
针对低能见度状态下对车辆与行人的视觉特征难以提取的问题,提出一种将2路卷积神经网络融合从而实现对车辆与行人识别的方法。采用高斯背景差分法实现图像去模糊,在双路网络中分别采用不同尺寸的滤波器,调整滤波器的大小得到不同环境下图片的特征值,采用反向传播算法计算梯度。实验结果显示,与单路式卷积神经网络对比,在能见度低的环境中,该方法对车辆的辨识率提高至83. 49%,对行人的辨识率提高至87. 36%,表明在低能见度环境中,双路式卷积神经网络识别准确率高于单路式卷积神经网络。
Aiming at the problem that it is difficult to extract the visual characteristics of vehicles and pedestrians under low visibility,a two-way convolution neural network fusion algorithm is proposed to identify vehicle and pedestrian. Using Gauss background difference method to remove fuzziness,and different sizes of filters are used in two-way networks. The size of the filter can be adjusted to get the eigenvalues of the images in different environments,and the backpropagation algorithm is used to calculate the gradient. The experimental results show thatcompared with the single-channel convolution neural network,the recognition rate of the proposed method is increased to 83. 49% for vehicles and 87. 36% for pedestrians in the low visibility environment,which indicates that the recognition accuracy of the two-way convolution neural network is higher than that of the single-way convolution neural networks.
Aiming at the problem that it is difficult to extract the visual characteristics of vehicles and pedestrians under low visibility,a two-way convolution neural network fusion algorithm is proposed to identify vehicle and pedestrian. Using Gauss background difference method to remove fuzziness,and different sizes of filters are used in two-way networks. The size of the filter can be adjusted to get the eigenvalues of the images in different environments,and the backpropagation algorithm is used to calculate the gradient. The experimental results show thatcompared with the single-channel convolution neural network,the recognition rate of the proposed method is increased to 83. 49% for vehicles and 87. 36% for pedestrians in the low visibility environment,which indicates that the recognition accuracy of the two-way convolution neural network is higher than that of the single-way convolution neural networks.
引文
[1]李琳辉,伦智梅,连静,等.基于卷积神经网络的道路车辆检测方法[J].吉林大学学报(工学版),2017,47(2):384-391.
[2]EIGEN D,PUHRSCH C,FERGUS R.Depth map prediction from a single image using a multi-scale deep network[J].MIT Press,2014:2366-2374.
[3]SIMONYAN K,ZISSERMAN A.Two-stream convolutional networks for action recognition in videos[J].Advances in Neural Information Processing Systems,2014,1(4):568-576.
[4]吴琼,陈锻生.多尺度卷积循环神经网络的情感分类技术[J].华侨大学学报(自然科学版),2017,38(6):875-879.
[5]高琳,王俊峰,范勇,等.基于卷积神经网络与一致性预测器的稳健视觉跟踪[J].光学学报,2017,37(8):229-238.
[6]袁智,胡辉.一种基于双路卷积神经网络跌倒识别方法[J].河南师范大学学报(自然版),2017(3):96-101.
[7]LIN T Y,ROYCHOWDHURY A,MAJI S.Bilinear CNN models for fine-grained visual recognition[J].IEEE International Conference on Computer Vision(ICCV),2015:1449-1457.
[8]任静静,方贤勇,陈尚文,等.基于快速卷积神经网络的图像去模糊[J].计算机辅助设计与图形学学报,2017,29(8):1444-1456.
[9]HUTTUNEN H,YANCHESHMEH F S,CHEN K.Car type recognition with Deep Neural Networks[C]//Intelligent Vehicles Symposium,June 19-22,2016;IV 2016 IEEE,Gotenburg.Sweden:IEEE,2016:1115-1120.
[10]TEICHMANN M,WEBER M,ZOELLNER M,et al.Multi Net:Real-time joint semantic reasoning for autonomous driving[J/OL].[2018-2-20].https://www.researchgate.net/publication/311842448_Multi Net_Real-time_Joint_Semantic_Reasoning_for_Autonomous_Driving.2016.
[11]吴群,王田,王汉武,等.现代智能视频监控研究综述[J].计算机应用研究,2016,33(6):1601-1606.
[12]邓柳,汪子杰.基于深度卷积神经网络的车型识别研究[J].计算机应用研究,2016,33(3):930-932.
[2]EIGEN D,PUHRSCH C,FERGUS R.Depth map prediction from a single image using a multi-scale deep network[J].MIT Press,2014:2366-2374.
[3]SIMONYAN K,ZISSERMAN A.Two-stream convolutional networks for action recognition in videos[J].Advances in Neural Information Processing Systems,2014,1(4):568-576.
[4]吴琼,陈锻生.多尺度卷积循环神经网络的情感分类技术[J].华侨大学学报(自然科学版),2017,38(6):875-879.
[5]高琳,王俊峰,范勇,等.基于卷积神经网络与一致性预测器的稳健视觉跟踪[J].光学学报,2017,37(8):229-238.
[6]袁智,胡辉.一种基于双路卷积神经网络跌倒识别方法[J].河南师范大学学报(自然版),2017(3):96-101.
[7]LIN T Y,ROYCHOWDHURY A,MAJI S.Bilinear CNN models for fine-grained visual recognition[J].IEEE International Conference on Computer Vision(ICCV),2015:1449-1457.
[8]任静静,方贤勇,陈尚文,等.基于快速卷积神经网络的图像去模糊[J].计算机辅助设计与图形学学报,2017,29(8):1444-1456.
[9]HUTTUNEN H,YANCHESHMEH F S,CHEN K.Car type recognition with Deep Neural Networks[C]//Intelligent Vehicles Symposium,June 19-22,2016;IV 2016 IEEE,Gotenburg.Sweden:IEEE,2016:1115-1120.
[10]TEICHMANN M,WEBER M,ZOELLNER M,et al.Multi Net:Real-time joint semantic reasoning for autonomous driving[J/OL].[2018-2-20].https://www.researchgate.net/publication/311842448_Multi Net_Real-time_Joint_Semantic_Reasoning_for_Autonomous_Driving.2016.
[11]吴群,王田,王汉武,等.现代智能视频监控研究综述[J].计算机应用研究,2016,33(6):1601-1606.
[12]邓柳,汪子杰.基于深度卷积神经网络的车型识别研究[J].计算机应用研究,2016,33(3):930-932.