融合全局选择与局部区分的车辆重识别网络

doi:10.16182/j.issn1004731x.joss.23-1009

摘要/Abstract

摘要：

针对跨镜头多视角差异导致车辆重识别面临的不同视角、复杂背景和光照强度等干扰问题，提出了一种融合全局选择与局部区分的车辆重识别网络。基于Resnet50骨干网络，设计了融合全局特征与局部特征的三分支互补网络，利用全局分支学习车辆的整体外观信息，局部分支捕获车辆的差异性细节信息。基于注意力机制提出了上下文特征选择模块（context feature selection module， CFSM），有效分离了车辆信息与复杂背景信息，并提出了一种细节特征增强模块（detail feature enhancement module，DFEM），利用部件之间的相对位置信息强化多粒度特征细节信息的学习。提出了一种权值自适应平衡策略，联合多损失函数进行训练。实验结果表明，所提网络在VeRi-776数据集上的mAP、CMC@1和CMC@5分别达到73.2%、93.4%和97.3%；在VehicleID数据集的大规模测试子集上，CMC@1和CMC@5分别达到75.0%和92.7%。与对比网络相比，所提网络具有较高的识别率和鲁棒性。

关键词: 车辆重识别, 多分支结构, 全局上下文特征, 局部区分特征, 权值自适应策略

Abstract:

To address the interference issues of different perspectives, complex backgrounds, and lighting intensity in vehicle re-identification caused by cross lens multi view differences, a vehicle re-identification network integrating global selection and local differentiation is proposed. Based on Resnet50 backbone network, a three-branch complementary network integrating global and local features is designed. The global branch is used to learn overall appearance information of the vehicle, while the local branch captures differential details of the vehicle. Based on attention mechanism, a context feature selection module (CFSM)is proposed to effectively separate vehicle information from complex background information, and a detail feature enhancement module (DFEM) is proposed to enhance the learning of multi granularity feature detail information by utilizing relative position information between components. A weight adaptive balancing strategyis proposed in combination with multiple loss functions for training. Experimental results show that on VeRi-776 dataset, the proposed network achieves 73.2%, 93.4% and 97.3% respectively on the mAP, CMC@1 and CMC@5. On a large-scale testing subset of VehicleID dataset, the proposed network achieves 75.0% and 92.7% respectively on the CMC@1 and CMC@5. Compared with the comparison network, the proposed network has higher recognition rate and robustness.

Key words: vehicle re-identification, multi-branch structure, global context features, local distinguishing features, weight adaptive strategy

中图分类号:

TP391.9

徐胜军,张梦倩,詹博涵等 . 融合全局选择与局部区分的车辆重识别网络[J]. 系统仿真学报, 2025, 37(1): 220-233.

Xu Shengjun,Zhang Mengqian,Zhan Bohan,et al . Global Selection and Local Differentiation Fusion for Vehicle Re-identification[J]. Journal of System Simulation, 2025, 37(1): 220-233.

图/表 15

图1

图2

图3

图4

图5

图6

表1

表2

表3

表4

权值自适应策略的最优选择 (%)

Method(Re-Rank)	mAP	CMC@1	CMC@5
(a) 权值参数为固定值2:1	75.97	94.28	95.71
(b) MALM的自适应权重	75.70	93.80	95.65
(c) 标准差 $σ$ 为阈值点， $ω + β = 1$	76.58	94.70	95.83
(d) 平均值 $x ¯$ 为阈值点， $ω + β = 1$	76.33	94.10	95.53
(e) 标准差 $σ$ 为阈值点， $2 ω + β 1 = 2$ 且 $ω + β 2 = 1$	75.86	93.86	95.77
(f) 平均值 $x ¯$ 为阈值点， $2 ω + β 1 = 2$ 且 $ω + β 2 = 1$	76.17	94.22	95.59

表4

表5

表6

图7

图8

图9

参考文献 33

1	Bai Yan, Gao Feng, Lou Yihang, et al. Incorporating Intra-class Variance to Fine-grained Visual Recognition[C]//2017 IEEE International Conference on Multimedia and Expo (ICME). Piscataway: IEEE, 2017: 1452-1457.
2	Jin Yi, Li Chenning, Li Yidong, et al. Model Latent Views with Multi-center Metric Learning for Vehicle Re-identification[J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 22(3): 1919-1931.
3	罗慧诚, 汪淑娟. 基于特征鲁棒性增强的多摄像头下车辆识别方法[J]. 系统仿真学报, 2023, 35(5): 1059-1074.
	Luo Huicheng, Wang Shujuan. Multi-camera Vehicle Recognition Method Based on Feature Robustness Enhancement[J]. Journal of System Simulation, 2023, 35(5): 1059-1074.
4	Wu Minghu, Qian Yeqiang, Wang Chunxiang, et al. A Multi-camera Vehicle Tracking System Based on City-scale Vehicle Re-ID and Spatial-temporal Information[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Piscataway: IEEE, 2021: 4072-4081.
5	Watcharapinchai Nattachai, Rujikietgumjorn Sitapa. Approximate License Plate String Matching for Vehicle Re-identification[C]//2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). Piscataway: IEEE, 2017: 1-6.
6	Khorramshahi P, Kumar A, Peri N, et al. A Dual-path Model with Adaptive Attention for Vehicle Re-identification[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2019: 6131-6140.
7	Tang Lisha, Wang Yi, Chau L P. Weakly-supervised Part-attention and Mentored Networks for Vehicle Re-identification[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(12): 8887-8898.
8	Sun Wei, Dai Guangzhao, Zhang Xiaorui, et al. TBE-net: A Three-branch Embedding Network with Part-aware Ability and Feature Complementary Learning for Vehicle Re-identification[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(9): 14557-14569.
9	Xu Yan, Rong Leilei, Zhou Xiaolei, et al. Joint Multiple Fine-grained feature for Vehicle Re-identification[J]. Array, 2022, 14: 100152.
10	Tu Mingfei, Zhu Kuan, Guo Haiyun, et al. Multi-granularity Mutual Learning Network for Object Re-identification[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(9): 15178-15189.
11	Sun Ziruo, Nie Xiushan, Bi Xiaopeng, et al. Detail Enhancement-based Vehicle Re-identification with Orientation-guided Re-ranking[J]. Pattern Recognition, 2023, 137: 109304.
12	Jiang Gangwu, Pang Xiyu, Tian Xin, et al. Global Reference Attention Network for Vehicle Re-identification[J]. Applied Intelligence, 2023, 53(9): 11328-11343.
13	Zheng Yanli, Pang Xiyu, Jiang Gangwu, et al. Dual-relational Attention Network for Vehicle Re-identification[J]. Applied Intelligence, 2023, 53(7): 7776-7787.
14	Tian Xin, Pang Xiyu, Jiang Gangwu, et al. Vehicle Re-identification Based on Global Relational Attention and Multi-granularity Feature Learning[J]. IEEE Access, 2022, 10: 17674-17682.
15	He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Deep Residual Learning for Image Recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 770-778.
16	Zhu Wenqian, Wang Zhongyuan, Wang Xiaochen, et al. A Dual Self-attention Mechanism for Vehicle Re-identification[J]. Pattern Recognition, 2023, 137: 109258.
17	Zhang Zhizheng, Lan Cuiling, Zeng Wenjun, et al. Relation-aware Global Attention for Person Re-identification[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 3183-3192.
18	Hermans Alexander, Beyer Lucas, Leibe Bastian. In Defense of the Triplet Loss for Person Re-identification[EB/OL]. (2017-11-21) [2023-07-16]. .
19	Su V Huynh, Nguyen Nam H, Nguyen Ngoc T, et al. A Strong Baseline for Vehicle Re-identification[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Piscataway: IEEE, 2021: 4142-4149.
20	Liu Xinchen, Liu Wu, Mei Tao, et al. PROVID: Progressive and Multimodal Vehicle Reidentification for Large-scale Urban Surveillance[J]. IEEE Transactions on Multimedia, 2018, 20(3): 645-658.
21	Liu Hongye, Tian Yonghong, Wang Yaowei, et al. Deep Relative Distance Learning: Tell the Difference Between Similar Vehicles[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 2167-2175.
22	Kingma Diederik P, Ba Lei. Adam: A Method for Stochastic Optimization[EB/OL]. (2017-01-30) [2023-07-16]. .
23	Zhong Zhun, Zheng Liang, Cao Donglin, et al. Re-ranking Person Re-identification with K-reciprocal Encoding[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2017: 3652-3661.
24	Lou Yihang, Bai Yan, Liu Jun, et al. Embedding Adversarial Learning for Vehicle Re-identification[J]. IEEE Transactions on Image Processing, 2019, 28(8): 3794-3807.
25	Liu Xiaobin, Zhang Shiliang, Huang Qingming, et al. RAM: A Region-aware Deep Model for Vehicle Re-identification[C]//2018 IEEE International Conference on Multimedia and Expo (ICME). Piscataway: IEEE, 2018: 1-6.
26	Kanacı Aytaç, Li Minxian, Gong Shaogang, et al. Multi-task Mutual Learning for Vehicle Re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. Piscataway: IEEE, 2019: 62-70.
27	Chu Ruihang, Sun Yifan, Li Yadong, et al. Vehicle Re-identification with Viewpoint-aware Metric Learning[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2019: 8281-8290.
28	Kuma R, Weill E, Aghdasi F, et al. Vehicle Re-identification: An Efficient Baseline Using Triplet Embedding[C]//2019 International Joint Conference on Neural Networks (IJCNN). Piscataway: IEEE, 2019: 1-9.
29	Teng Shangzhi, Zhang Shiliang, Huang Qingming, et al. Viewpoint and Scale Consistency Reinforcement for UAV Vehicle Re-identification[J]. International Journal of Computer Vision, 2021, 129(3): 719-735.
30	Yao Aihuan, Huang Mengmeng, Qi Jiahao, et al. Attention Mask-based Network with Simple Color Annotation for UAV Vehicle Re-identification[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19: 1-5.
31	Guo Haiyun, Zhao Chaoyang, Liu Zhiwei, et al. Learning Coarse-to-fine Structured Feature Embedding for Vehicle Re-identification[C]//Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence and Thirtieth Innovative Applications of Artificial Intelligence Conference and Eighth AAAI Symposium on Educational Advances in Artificial Intelligence. Palo Alto: AAAI Press, 2018: 6853-6860.
32	He Bing, Li Jia, Zhao Yifan, et al. Part-regularized Near-duplicate Vehicle Re-identification[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2019: 3992-4000.
33	Qian Jiuchao, Pan Minting, Tong Wei, et al. URRNet: A Unified Relational Reasoning Network for Vehicle Re-identification[J]. IEEE Transactions on Vehicular Technology, 2023, 72(9): 11156-11168.

Method (Re-Rank)	CMC@5	CMC@10
(a) 竖直4，竖直2	94.82	96.60
(b) 竖直4	93.56	96.31
(c)竖直2	94.76	96.72
(d)竖直2，水平2	95.35	97.20
(e)竖直4，水平2	94.70	96.96
(f)竖直4，竖直4，水平2	94.82	96.66

Method (Re-Rank)	mAP	CMC@1	CMC@5
(a)单分支输出	72.44	91.84	94.22
(b)双分支输出	76.06	93.45	95.29

Method (Re-Rank)	mAP	CMC@1	CMC@5
Baseline	76.60	93.62	95.59
Baseline +CFSM	76.29	93.38	95.59
Baseline+CFSM +1*1Conv	76.86	94.10	96.01
Baseline + DFEM(=Lobal-1)	76.63	94.04	95.65
Baseline + DFEM (=Lobal-2)	76.74	93.80	95.77
Baseline + DFEM (=Lobal-1+ Lobal-2)	76.56	93.98	95.83
GSLD (ours)	77.17	94.28	96.35

Method(w/o Re-Rank)	mAP	CMC@1	CMC@5
EALN	57.4	84.4	94.1
RAM	61.5	88.6	94.0
AAVER	61.2	89.0	94.7
MTML	64.6	92.3	95.7
VANet	66.2	89.2	95.9
Batch sample	67.6	90.2	96.4
MSA	72.9	92.1	96.2
AM	71.1	93.5	97.1
GSLD (ours)	73.2	93.4	97.3

Method(w/o Re-Rank)	Test800		Test1600		Test2400
Method(w/o Re-Rank)	CMC@1	CMC@5	CMC@1	CMC@5	CMC@1	CMC@5
C2F-Rank	61.1	81.7	56.2	76.2	51.4	72.2
RAM	75.2	91.5	72.3	87.0	67.7	84.5
AAVER	74.7	93.8	68.6	90.0	63.5	85.6
PRN	78.4	92.3	75.0	88.3	74.2	86.4
MSA	77.6	90.5	74.4	86.3	72.9	84.6
URRNet	76.5	96.5	73.7	92.0	68.2	89.6
DSAM	80.6	94.8	78.2	92.6	75.0	89.2
GRMF	81.5	96.3	76.8	93.6	72.8	91.0
GSLD (ours)	82.6	97.8	77.5	94.8	75.0	92.7