Faster R-CNN模型在车辆检测中的应用
|
摘要
针对传统机器学习方法在车辆检测应用中易受光照、目标尺度和图像质量等因素影响,效率低下且泛化能力较差的问题,提出一种基于改进的较快的基于区域卷积神经网络(R-CNN)模型的车辆检测方法。该方法以Faster R-CNN模型为基础,通过对输入图像进行卷积和池化等操作提取车辆特征,结合多尺度训练和难负样本挖掘策略降低复杂环境的影响,利用KITTI数据集对深度神经网络模型进行训练,并采集实际场景中的图像进行测试。仿真实验中,在保证检测时间的情况下,相对原Faster R-CNN算法检测精确度提高了约8%。实验结果表明,所提方法能够自动地提取车辆特征,解决了传统方法提取特征费时费力的问题,同时提高了车辆检测精确度,具有良好的泛化能力和适用范围。
Since the traditional machine learning methods are easy to be affected by light, target scale and image quality in vehicle detection applications, resulting the low efficiency and generalization ability, a vehicle detection method based on improved Faster Regions with Convolutional Neural Network features(R-CNN) model was proposed. On the basis of Faster RCNN model, through convolution and pooling operations to extract the features of vehicles, by combining with multi-scale training and hard negative sample mining strategy to reduce the influence of complex environment, the KITTI data set was used to train the deep neural network model, and the images were collected from actual scene to test the trained neural network model. In the simulation experiments, while the detection time was guaranteed, the detection accuracy of the proposed method was improved by about 8% compared to the original Faster R-CNN algorithm. The experimental results show that the proposed method can automatically extract the features of vehicles, solve the time-consuming and laborious problem of extracting features by traditional methods, effectively improve the accuracy of vehicle detection, and has good generalization ability and wide range of applications.
Since the traditional machine learning methods are easy to be affected by light, target scale and image quality in vehicle detection applications, resulting the low efficiency and generalization ability, a vehicle detection method based on improved Faster Regions with Convolutional Neural Network features(R-CNN) model was proposed. On the basis of Faster RCNN model, through convolution and pooling operations to extract the features of vehicles, by combining with multi-scale training and hard negative sample mining strategy to reduce the influence of complex environment, the KITTI data set was used to train the deep neural network model, and the images were collected from actual scene to test the trained neural network model. In the simulation experiments, while the detection time was guaranteed, the detection accuracy of the proposed method was improved by about 8% compared to the original Faster R-CNN algorithm. The experimental results show that the proposed method can automatically extract the features of vehicles, solve the time-consuming and laborious problem of extracting features by traditional methods, effectively improve the accuracy of vehicle detection, and has good generalization ability and wide range of applications.
引文
[1]NEGRI P,CLADY X,HANIF S M,et al.A cascade of boosted generative and discriminative classifiers for vehicle detection[J].Eurasip Journal on Advances in Signal Processing,2008,2008:Article No.136.
[2]MA X,GRIMSON W E L.Edge-based rich representation for vehicle classification[C]//Proceedings of the 2005 10th IEEE International Conference on Computer Vision.Washington,DC:IEEE Computer Society,2005:1185-1192.
[3]TEOH S S,BRAUNL T.Symmetry-based monocular vehicle detection system[J].Machine Vision&Applications,2012,23(5):831-842.
[4]CAO X,WU C,YAN P,et al.Linear SVM classification using boosting HOG features for vehicle detection in low-altitude airborne videos[C]//Proceedings of the 2011 IEEE International Conference on Image Processing.Piscataway,NJ:IEEE,2011:2421-2424.
[5]何灼彬.基于卷积深度置信网络的歌手识别[D].广州:华南理工大学,2015:38-48.(HE Z B.Singer identification based on convolution deep belief networks[D].Guangzhou:South China University of Technology,2015:38-48.)
[6]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2014:580-587.
[7]HE K,ZHANG X,REN S,et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2015,37(9):1904-1916.
[8]GIRSHICK R.Fast R-CNN[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2015:1440-1448.
[9]REN S,HE K,GIRSHICK R,et al.Faster R-CNN:towards realtime object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2017,39(6):1137-1149.
[10]SUN X,WU P,HOI S C H.Face detection using deep learning:an improved faster RCNN approach[EB/OL].[2017-03-01].http://xueshu.baidu.com/s?wd=paperuri%3A%28526a86e5697d1f0c149d60d8ba856dd5%29&filter=sc-longsign&tn=SE-xueshusource-2kduw22v&sc-vurl=http%3A%2F%2Farxiv.org%2Fpdf%2F1701.08289&ie=utf-8&sc-us=9408430938623955412.
[11]SHRIVASTAVA A,GUPTA A,GIRSHICK R.Training regionbased object detectors with online hard example mining[C]//CVPR 2016:Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2016:761-769.
[12]FAN Q,BROWN L,SMITH J.A closer look at faster R-CNN for vehicle detection[C]//Proceedings of the 2016 Intelligent Vehicles Symposium.Piscataway,NJ:IEEE,2016:124-129.
[13]钟晓明,余贵珍,马亚龙,等.基于快速区域卷积神经网络的交通标志识别算法研究[C]//中国汽车工程学会年会论文集.北京:中国汽车工程学会,2016:2033-2036.(ZHONG X M,YU G Z,MA Y L,et al.Research on traffic sign recognition algorithm based on faster R-CNN[C]//Proceedings of the 2016 Annual Conference of Society of Automotive Engineers of China.Beijing:SAE-China,2016:2033-2036.)
[14]REDMON J,DIVVALA S,GIRSHICK R,et al.You only look once:unified,real-time object detection[C]//CVPR 2016:Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2016:779-788.
[15]LIU W,ANGUELOV D,ERHAN D,et al.SSD:single shot multibox detector[C]//Proceedings of the 2016 European Conference on Computer Vision,LNCS 9905.Berlin:Springer,2016:21-37.
[16]DAI J,LI Y,HE K,et al.R-FCN:object detection via region-based fully convolutional networks[EB/OL].[2017-04-21].http://pdfs.semanticscholar.org/a8f2/4fcc1eb0354ffd91f0e3031f5c4dc3e02dd6.pdf.
[2]MA X,GRIMSON W E L.Edge-based rich representation for vehicle classification[C]//Proceedings of the 2005 10th IEEE International Conference on Computer Vision.Washington,DC:IEEE Computer Society,2005:1185-1192.
[3]TEOH S S,BRAUNL T.Symmetry-based monocular vehicle detection system[J].Machine Vision&Applications,2012,23(5):831-842.
[4]CAO X,WU C,YAN P,et al.Linear SVM classification using boosting HOG features for vehicle detection in low-altitude airborne videos[C]//Proceedings of the 2011 IEEE International Conference on Image Processing.Piscataway,NJ:IEEE,2011:2421-2424.
[5]何灼彬.基于卷积深度置信网络的歌手识别[D].广州:华南理工大学,2015:38-48.(HE Z B.Singer identification based on convolution deep belief networks[D].Guangzhou:South China University of Technology,2015:38-48.)
[6]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2014:580-587.
[7]HE K,ZHANG X,REN S,et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2015,37(9):1904-1916.
[8]GIRSHICK R.Fast R-CNN[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2015:1440-1448.
[9]REN S,HE K,GIRSHICK R,et al.Faster R-CNN:towards realtime object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2017,39(6):1137-1149.
[10]SUN X,WU P,HOI S C H.Face detection using deep learning:an improved faster RCNN approach[EB/OL].[2017-03-01].http://xueshu.baidu.com/s?wd=paperuri%3A%28526a86e5697d1f0c149d60d8ba856dd5%29&filter=sc-longsign&tn=SE-xueshusource-2kduw22v&sc-vurl=http%3A%2F%2Farxiv.org%2Fpdf%2F1701.08289&ie=utf-8&sc-us=9408430938623955412.
[11]SHRIVASTAVA A,GUPTA A,GIRSHICK R.Training regionbased object detectors with online hard example mining[C]//CVPR 2016:Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2016:761-769.
[12]FAN Q,BROWN L,SMITH J.A closer look at faster R-CNN for vehicle detection[C]//Proceedings of the 2016 Intelligent Vehicles Symposium.Piscataway,NJ:IEEE,2016:124-129.
[13]钟晓明,余贵珍,马亚龙,等.基于快速区域卷积神经网络的交通标志识别算法研究[C]//中国汽车工程学会年会论文集.北京:中国汽车工程学会,2016:2033-2036.(ZHONG X M,YU G Z,MA Y L,et al.Research on traffic sign recognition algorithm based on faster R-CNN[C]//Proceedings of the 2016 Annual Conference of Society of Automotive Engineers of China.Beijing:SAE-China,2016:2033-2036.)
[14]REDMON J,DIVVALA S,GIRSHICK R,et al.You only look once:unified,real-time object detection[C]//CVPR 2016:Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington,DC:IEEE Computer Society,2016:779-788.
[15]LIU W,ANGUELOV D,ERHAN D,et al.SSD:single shot multibox detector[C]//Proceedings of the 2016 European Conference on Computer Vision,LNCS 9905.Berlin:Springer,2016:21-37.
[16]DAI J,LI Y,HE K,et al.R-FCN:object detection via region-based fully convolutional networks[EB/OL].[2017-04-21].http://pdfs.semanticscholar.org/a8f2/4fcc1eb0354ffd91f0e3031f5c4dc3e02dd6.pdf.