鉴别性特征学习模型实现跨摄像头下行人即时对齐
|
摘要
为解决由于采用延后的关联算法而造成目标错误匹配和子序列漏匹配的问题,提出一种使用鉴别性特征学习模型实现跨摄像头下行人即时对齐的方法.首先基于孪生网络模型整合行人分类和行人身份鉴别模型,仅通过目标行人的单帧信息就可习得具有良好鉴别性的行人外观特征,完成行人相似性值计算;其次提出跨摄像头行人即时对齐模型,根据行人外观、时序和空间3个方面的关联适配度实时建立最小费用流图并求解.实验结果表明,在行人重识别数据集Market-1501和CUHK03上,行人分类和身份鉴别模型的融合能显著提升特征提取的有效性且泛化能力良好,性能全面优于Gate-SCNN与S-LSTM方法;进一步地,在非重叠区域的跨摄像头行人跟踪的基准数据集NLPR_MCT上,该方法的行人即时关联精度比2014年ECCV跨摄像头行人跟踪冠军的延后关联算法高出了3.3%,仅次于当前最高精度算法6.6%,应用于跨摄像头跟踪时,跟踪精度亦超过当前的大部分算法.
Those existing delayed association algorithms of non-overlapping cameras always lead to target mismatching and subsequence match missing. To solve these two problems, this paper proposes pedestrian instant alignment through discriminative feature learning. First, this paper employs a siamese network to integrates a pedestrian verification model with a pedestrian identification one. The integrated model can extract discriminative appearance features from single frame of targeted pedestrian and calculate similarity for a pair of pedestrians.Second, this paper presents an instant alignment model for pedestrians across non-overlapping cameras. The fundamental of proposed instant alignment model is minimum cost flow algorithm. Hence, according to a match degree which is associated with appearance, spatial and temporal context, a dynamic minimum cost flow graph is established and solved in real time. The experimental results show that, on the pedestrian recognition datasets Market-1501 and CUHK03, the combination of pedestrian verification and identification model can improve the efficiency of feature extraction and the generalization ability significantly. The alignment performance of the proposed model is superior to the Gate-SCNN and S-LSTM. Furthermore, on dataset NLPR_MCT, the benchmark dataset for pedestrian tracking of non-overlapping cameras, the instant alignment accuracy of the proposed model increases by 3.3% compared to the champion algorithm in 2014 ECCV cross-camera pedestrian tracking challenge, which is a delayed association algorithm. The experiment results also show that the proposed model ranks second, just 6.6% lower than the state-of-the-art performance. When the propose model is applied to inter-cameras pedestrian tracking, the tracking accuracy is also higher than most popular algorithms.
Those existing delayed association algorithms of non-overlapping cameras always lead to target mismatching and subsequence match missing. To solve these two problems, this paper proposes pedestrian instant alignment through discriminative feature learning. First, this paper employs a siamese network to integrates a pedestrian verification model with a pedestrian identification one. The integrated model can extract discriminative appearance features from single frame of targeted pedestrian and calculate similarity for a pair of pedestrians.Second, this paper presents an instant alignment model for pedestrians across non-overlapping cameras. The fundamental of proposed instant alignment model is minimum cost flow algorithm. Hence, according to a match degree which is associated with appearance, spatial and temporal context, a dynamic minimum cost flow graph is established and solved in real time. The experimental results show that, on the pedestrian recognition datasets Market-1501 and CUHK03, the combination of pedestrian verification and identification model can improve the efficiency of feature extraction and the generalization ability significantly. The alignment performance of the proposed model is superior to the Gate-SCNN and S-LSTM. Furthermore, on dataset NLPR_MCT, the benchmark dataset for pedestrian tracking of non-overlapping cameras, the instant alignment accuracy of the proposed model increases by 3.3% compared to the champion algorithm in 2014 ECCV cross-camera pedestrian tracking challenge, which is a delayed association algorithm. The experiment results also show that the proposed model ranks second, just 6.6% lower than the state-of-the-art performance. When the propose model is applied to inter-cameras pedestrian tracking, the tracking accuracy is also higher than most popular algorithms.
引文
[1]Wang X G.Intelligent multi-camera video surveillance:a review[J].Pattern Recognition Letters,2013,34(1):3-19
[2]Huang K Q,Tan T N.Vs-star:a visual interpretation system for visual surveillance[J].Pattern Recognition Letters,2010,31(14):2265-2285
[3]Huang C,Wu B,Nevatia R.Robust object tracking by hierarchical association of detection responses[C]//Proceedings of European Conference on Computer Vision.Heidelberg:Springer,2008:788-801
[4]Chen X J,Bhanu B.Integrating social grouping for multi-target tracking across cameras in a CRF model[J].IEEE Transactions on Circuits and Systems for Video Technology,2017,27(11):2382-2394
[5]Chen W H,Cao L J,Chen X T,et al.An equalized global graph model-based approach for multi-camera object tracking[J].IEEE Transactions on Circuits and Systems for Video Technology,2017,27(11):2367-2381
[6]Cai Y H,Medioni G.Exploring context information for inter-camera multiple target tracking[C]//Proceedings of the IEEE Winter Conference on Applications of Computer Vision.Los Alamitos:IEEE Computer Society Press,2014:761-768
[7]Chen W H,Cao L J,Chen X T,et al.A novel solution for multi-camera object tracking[C]//Proceedings of the IEEE International Conference on Image Processing.Los Alamitos:IEEE Computer Society Press,2014:2329-2333
[8]Kuo C H,Huang C,Nevatia R.Inter-camera association of multi-target tracks by on-line learned appearance affinity models[C]//Proceedings of the 11th European Conference on Computer Vision.Heidelberg:Springer,2010,Part I:383-396
[9]Cheng D,Gong Y H,Wang J J,et al.Part-aware trajectories association across non-overlapping uncalibrated cameras[J].Neurocomputing,2017,230(22):30-39
[10]Varior R R,Shuai B,Lu J W,et al.A siamese long short-term memory architecture for human re-identification[C]//Proceedings of the 14th European Conference on Computer Vision.Heidelberg:Springer,2016:135-153
[11]Varior R R,Haloi M,Wang G.Gated siamese convolutional neural network architecture for human re-identification[C]//Proceedings of the 14th European Conference on Computer Vision.Heidelberg:Springer,2016:791-808
[12]Hermans A,Beyer L,Leibe B.In defense of the triplet loss for person re-identification[OL].[2018-05-16].https://arxiv.org/abs/1703.07737
[13]Zheng L,Shen L Y,Tian L,et al.Scalable person re-identification:a benchmark[C]//Proceedings of the IEEEInternational Conference on Computer Vision.Los Alamitos:IEEE Computer Society Press,2015:1116-1124
[14]Li W,Zhao R,Xiao T,et al.DeepReID:deep filter pairing neural network for person re-identification[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2014:152-159
[15]Yi D,Lei Z,Liao S C,et al.Deep metric learning for person re-identification[C]//Proceedings of the 22nd International Conference on Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2014:34-39
[16]Liu Shaohua,Lai Shiming,Zhang Maojun.A min-cost flow based algorithm for object association of multiple non-overlappig cameras[J].Acta Automatica Sinca,2010,36(10):1484-1489(in Chinese)(刘少华,赖世铭,张茂军.基于最小费用流模型的无重叠视域多摄像机目标关联算法[J].自动化学报,2010,36(10):1484-1489)
[17]Lee Y G,Tang Z,Hwang J N.Online-learning-based human tracking across non-overlapping cameras[J].IEEE Transactions on Circuits and Systems for Video Technology,2018,28(10):2870-2883
[18]Ahmed E,Jones M,Marks T K.An improved deep learning architecture for person re-identification[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2015:3908-3916
[19]Hadsell R,Chopra S,LeCun Y.Dimensionality reduction by learning an invariant mapping[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2006,2:1735-1742
[20]Zheng Z D,Zheng L,Yang Y.A discriminatively learned CNNembedding for person reidentification[J].ACM Transactions on Multimedia Computing,Communications,and Applications,2018,14(1):Article No.13
[21]He K M,Zhang X Y,Ren S Q,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEEComputer Society Press,2016:770-778
[22]Henriques J F,Caseiro R,Martins P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596
[23]Liu W,Anguelov D,Erhan D,et al.SSD:single shot multibox detector[C]//Proceedings of European Conference on Computer Vision.Heidelberg:Springer,2016:21-37
[24]Barnich O,van Droogenbroeck M.ViBe:a universal background subtraction algorithm for video sequences[J].IEEETransactions on Image Processing,2011,20(6):1709-1724
[2]Huang K Q,Tan T N.Vs-star:a visual interpretation system for visual surveillance[J].Pattern Recognition Letters,2010,31(14):2265-2285
[3]Huang C,Wu B,Nevatia R.Robust object tracking by hierarchical association of detection responses[C]//Proceedings of European Conference on Computer Vision.Heidelberg:Springer,2008:788-801
[4]Chen X J,Bhanu B.Integrating social grouping for multi-target tracking across cameras in a CRF model[J].IEEE Transactions on Circuits and Systems for Video Technology,2017,27(11):2382-2394
[5]Chen W H,Cao L J,Chen X T,et al.An equalized global graph model-based approach for multi-camera object tracking[J].IEEE Transactions on Circuits and Systems for Video Technology,2017,27(11):2367-2381
[6]Cai Y H,Medioni G.Exploring context information for inter-camera multiple target tracking[C]//Proceedings of the IEEE Winter Conference on Applications of Computer Vision.Los Alamitos:IEEE Computer Society Press,2014:761-768
[7]Chen W H,Cao L J,Chen X T,et al.A novel solution for multi-camera object tracking[C]//Proceedings of the IEEE International Conference on Image Processing.Los Alamitos:IEEE Computer Society Press,2014:2329-2333
[8]Kuo C H,Huang C,Nevatia R.Inter-camera association of multi-target tracks by on-line learned appearance affinity models[C]//Proceedings of the 11th European Conference on Computer Vision.Heidelberg:Springer,2010,Part I:383-396
[9]Cheng D,Gong Y H,Wang J J,et al.Part-aware trajectories association across non-overlapping uncalibrated cameras[J].Neurocomputing,2017,230(22):30-39
[10]Varior R R,Shuai B,Lu J W,et al.A siamese long short-term memory architecture for human re-identification[C]//Proceedings of the 14th European Conference on Computer Vision.Heidelberg:Springer,2016:135-153
[11]Varior R R,Haloi M,Wang G.Gated siamese convolutional neural network architecture for human re-identification[C]//Proceedings of the 14th European Conference on Computer Vision.Heidelberg:Springer,2016:791-808
[12]Hermans A,Beyer L,Leibe B.In defense of the triplet loss for person re-identification[OL].[2018-05-16].https://arxiv.org/abs/1703.07737
[13]Zheng L,Shen L Y,Tian L,et al.Scalable person re-identification:a benchmark[C]//Proceedings of the IEEEInternational Conference on Computer Vision.Los Alamitos:IEEE Computer Society Press,2015:1116-1124
[14]Li W,Zhao R,Xiao T,et al.DeepReID:deep filter pairing neural network for person re-identification[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2014:152-159
[15]Yi D,Lei Z,Liao S C,et al.Deep metric learning for person re-identification[C]//Proceedings of the 22nd International Conference on Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2014:34-39
[16]Liu Shaohua,Lai Shiming,Zhang Maojun.A min-cost flow based algorithm for object association of multiple non-overlappig cameras[J].Acta Automatica Sinca,2010,36(10):1484-1489(in Chinese)(刘少华,赖世铭,张茂军.基于最小费用流模型的无重叠视域多摄像机目标关联算法[J].自动化学报,2010,36(10):1484-1489)
[17]Lee Y G,Tang Z,Hwang J N.Online-learning-based human tracking across non-overlapping cameras[J].IEEE Transactions on Circuits and Systems for Video Technology,2018,28(10):2870-2883
[18]Ahmed E,Jones M,Marks T K.An improved deep learning architecture for person re-identification[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2015:3908-3916
[19]Hadsell R,Chopra S,LeCun Y.Dimensionality reduction by learning an invariant mapping[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2006,2:1735-1742
[20]Zheng Z D,Zheng L,Yang Y.A discriminatively learned CNNembedding for person reidentification[J].ACM Transactions on Multimedia Computing,Communications,and Applications,2018,14(1):Article No.13
[21]He K M,Zhang X Y,Ren S Q,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEEComputer Society Press,2016:770-778
[22]Henriques J F,Caseiro R,Martins P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596
[23]Liu W,Anguelov D,Erhan D,et al.SSD:single shot multibox detector[C]//Proceedings of European Conference on Computer Vision.Heidelberg:Springer,2016:21-37
[24]Barnich O,van Droogenbroeck M.ViBe:a universal background subtraction algorithm for video sequences[J].IEEETransactions on Image Processing,2011,20(6):1709-1724