In pedestrian re-identification (Re-ID) tasks, insufficient feature discriminability, poor training stability, and weak robustness are often caused by viewpoint changes, occlusion, and background interference. A multi-strategy pedestrian re-identification network with enhanced robustness was proposed. Triplet Attention modules were embedded in the four stages of the ResNet50 backbone network. Channel, horizontal, and vertical attention mechanisms were used. Feature weights were dynamically adjusted. The focusing ability on key pedestrian areas was enhanced. Background noise was suppressed. A constraint term on the ratio of positive and negative sample distances was introduced on the basis of traditional Triplet Loss. The feature space distribution was finely regulated. Intra-class compactness was improved. Inter-class discriminability was enhanced. Kaiming initialization and normalization layer optimization strategies were adopted. Gradient instability in the early training stage was reduced. Network convergence was accelerated. Feature representation ability was enhanced. The mAP of the method is 88.4%, 78.9%, and 54.6% on the Market1501, DukeMTMC-ReID, and MSMT17_V2 datasets. The Rank-1 index is 95.3%, 88.9%, and 78.1%. These results are superior to those of existing mainstream methods. The visualization of heat maps and ranking charts shows that pedestrian details are accurately captured. Strong adaptability to complex scenes is achieved. This method provides an effective solution to improve the robustness of cross-camera pedestrian matching.
LIU Zi-liang, YAERMAIMAITI Yilihamu.
Robust Pedestrian Re-identification with Multi-dimensional Attention and Distance Ratio Constraint[J].
Science Technology and Engineering, 2026, 26(13): 5616-5627 DOI:10.12404/j.issn.1671-1815.2503711
综上,采用Kaiming初始化的初始卷积层与常数初始化的归一化层共同作用,显著增强了模型训练初期的稳定性,加速了收敛过程,并在多个Re-ID基准数据集上带来了Rank-1以及平均精度均值(mean average precision,mAP)的提升,为深层卷积结构在行人重识别等小样本、高变化场景下的有效迁移提供了理论与实践支撑。
热力图能够直观展示不同深度学习模型在行人图像上的注意力分布,有助于分析模型的特征提取能力。通过原图对基线模型BoT(bag of tricks)与改进算法的热力图,可以观察它们对行人不同部位的关注程度,了解模型是否专注于关键特征或受到背景干扰。如图6所示,第1列为原始图像,第2列为基线模型BoT,第3列为本文算法生成的热力图。
DasA, ChakrabortyA, Roy-ChowdhuryA K. Consistent re-identification in a camera network[C]// Proceedings of the European Conference on Computer Vision. Zurich: European Association for Computer Vision, 2014: 330-345.
[2]
ZhouX, CuiN, LiZ, et al. Hierarchical Gaussianization for image classification[C]// Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. Miami Beach: IEEE, 2009: 1971-1977.
[3]
YangL, WangY, LiuL, et al. Center prediction loss for re-identification[J]. Pattern Recognition, 2022, 132: 108949.
[4]
ZhouX, ZhongY, ChengZ, et al. Adaptive sparse pairwise loss for object re-identification[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 19691-19701.
[5]
WangG, YuanY, ChenX, et al. Learning discriminative features with multiple granularities for person re-identification[C]// Proceedings of the 26th ACM International Conference on Multimedia. Seoul: IEEE, 2018: 274-282
[6]
ZhangX, FuK, ZhaoQ. Dynamic patch-aware enrichment transformer for occluded person re-identification[J]. Knowledge-Based Systems, 2025, 327: 114193.
ChenYu, LiuHui, LiangDongsheng, et al. Person re-identification based on pose estimation and Transformer model[J]. Science Technology and Engineering, 2024, 24(12): 5051-5058.
QinPeng, ChenGaohua, GuJiaxin. Occluded pedestrian re-identification based on foreground segmentation and multi-scale feature fusion[J]. Science Technology and Engineering, 2025, 25(21): 9002-9009.
[11]
LuoH, GuY, LiaoX, et al. Bag of tricks and a strong baseline for deep person re-identification[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Long Beach: IEEE, 2019: 1487-1495.
[12]
MisraD, NalamadaT, ArasanipalaiA U, et al. Rotate to attend: convolutional triplet attention module[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. Waikoloa: IEEE, 2021: 3139-3148.
[13]
HeK, ZhangX, RenS, et al. Delving deep into rectifiers: surpassing human-level performance on imagenet classification[C]// Proceedings of the IEEE International Conference on Computer Vision. Santiago: IEEE, 2015: 1026-1034.
[14]
ZhengL, ShenL, TianL, et al. Scalable person re-identification: a benchmark[C]// Proceedings of the IEEE International Conference on Computer Vision. Santiago: IEEE, 2015: 1116-1124.
[15]
ZhengZ, ZhengL, YangY. Unlabeled samples generated by gan improve the person re-identification baseline in vitro[C]// Proceedings of the IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 3754-3762.
[16]
WeiL, ZhangS, GaoW, et al. Person transfer GAN to bridge domain gap for person re-identification[C]// Proceedings of the IEEE Conference on computer vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 79-88.
[17]
GeY, LiZ, ZhaoH, et al. FD-GAN: pose-guided feature distilling GAN for robust person re-identification[J]. Advances in Neural Information Processing Systems, 2018, 31: 1230-1241.
[18]
SunY, ZhengL, YangY, et al. Beyond part models: person retrieval with refined part pooling (and a strong convolutional baseline)[C]// Proceedings of the European Conference on Computer Vision (ECCV). Munich: ECCV, 2018: 480-496.
[19]
ZhouK, YangY, CavallaroA, et al. Omni-scale feature learning for person re-identification[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 3702-3712.
[20]
WangG, YangS, LiuH, et al. High-order information matters: learning relation and topology for occluded person re-identification[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Online: IEEE, 2020: 6449-6458.
[21]
YeM, ShenJ, LinG, et al. Deep learning for person re-identification: a survey and outlook[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(6): 2872-2893.
[22]
HanX, YuX, LiG, et al. Rethinking sampling strategies for unsupervised person re-identification[J]. IEEE Transactions on Image Processing, 2022, 32: 29-42.
[23]
JiaM, ChengX, LuS, et al. Learning disentangled representation implicitly via transformer for occluded person re-identification[J]. IEEE Transactions on Multimedia, 2022, 25: 1294-1305.
[24]
HeQ, WangZ, ZhengZ, et al. Spatial and temporal dual-attention for unsupervised person re-identification[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 25(2): 1953-1965.
PengJinjia, WangHuibing. An unsupervised person re-identification method based on heterogeneous convolutional neural networks ensemble[J]. Acta Electronica Sinica, 2023, 51(10): 2902-2914.
[27]
WangJ, WangJ. MHDNet: a multi-scale hybrid deep learning model for person re-identification[J]. Electronics, 2024, 13(8): 1435.
[28]
GongY, ZhangC, HouY, et al. Beyond dropout: robust convolutional neural networks based on local feature masking[C]// 2024 International Joint Conference on Neural Networks (IJCNN). Pacifico Yokohama: IEEE, 2024: 1-8.
[29]
JiH, WangL, ZhouS, et al. Transfer easy to hard: adversarial contrastive feature learning for unsupervised person re-identification[J]. Pattern Recognition, 2024, 145: 109973.
JiaJieru, ZhangShuorui, QianYuhua, et al. Unsupervised pedestrian reidentification based on pseudo-label regularization loss[J]. Journal of Electronics, 2024, 52(5): 1743-1758.