College of Information and Cyber Security, People’s Public Security University of China, Beijing 100038, China
Show less
文章历史+
Received
Published
Revised
2025-08-10
2026-07-08
2025-12-16
Online Date
2026-08-06
PDF (6433K)
摘要
在弱监督异常行为检测中,从视频级标签视频中选择正常与异常关键实例,以学习异常事件与正常状态的判别特征至关重要。现有的特征选择策略主要依据单个实例的异常强度,忽略了实例间的全局统计特性。为此,提出一种联合记忆与均值统计增强的特征选择方法。该方法通过存储全局正常模式与均值统计量作为锚点,计算输入视频特征相对于锚点的偏离度,依据偏离度大小选择正样本与负样本,并构建三元组损失函数进行约束,通过关键样本的区分增强正常与异常判别特征。此外,鉴于正常与异常视频片段的边界具有模糊性,设计了滑动窗口约束的动态图卷积模块,增强对渐变过程的特征表示和对局部异常模式的关注。实验结果表明,该方法在UCF-Crime和XD-Violence数据集上的AUC(area under curve)和AP(average precision)指标分别达到86.25%和83.87%。可见,提出的基于记忆与均值两种全局统计量的特征选择策略,能够有效筛选关键样本实例,显著增强模型对正常与异常模式的判别特征表示能力。
Abstract
In weakly supervised anomaly behavior detection, selecting critical normal and anomalous instances from video-level labeled videos is essential for learning discriminative features that distinguish abnormal events from normal states. Existing feature selection strategies are primarily based on the anomaly intensity of individual instances, while the global statistical characteristics among instances are ignored. To address this, a joint memory and mean statistics enhanced feature selection method was proposed. In this method, global normal patterns and mean statistics were stored as anchor points, the deviation degree of input video features relative to these anchors was calculated, and positive and negative samples were selected based on the deviation magnitude. Then, a triplet loss function was constructed to constrain the samples, and the discriminative features between normal and abnormal states were enhanced through the discrimination of key samples. Furthermore, considering the ambiguity of boundaries between normal and abnormal video segments, a dynamic graph convolution module with sliding window constraints was designed to strengthen the feature representation of gradient processes and focus on local abnormal patterns. The experimental results demonstrate that AUC(area under curve) and AP(average precision) metrics of 86.25% and 83.87% are achieved by the proposed method on the UCF-Crime and XD-Violence datasets, respectively, verifying its effectiveness. It is proved that the feature selection strategy based on two global statistics of memory and mean proposed can effectively select key sample instances and significantly enhance the ability of the model to discriminate between normal and abnormal patterns.
在深度学习领域,模型性能提升常依赖加深网络层数,但这易引发计算量激增与训练效率下降问题。为在不增加网络深度的前提下增强模型特征提取能力,同时解决传统卷积方法在计算资源与效率上的局限,引入动态分组卷积[14](dynamic group convolution, DGC)替代依赖叠加层深的卷积方法,以降低计算消耗并提升效率。该方法的每层卷积并非共享单一卷积核,而是基于动态生成的注意力权重聚合多个并行卷积核。
WuP, PanC Y, YanY T, et al. Deep learning for video anomaly detection: a review[C]// IEEE Transactions on Neural Networks and Learning Systems(TNNLS). New York: IEEE, 2024: 3647892.
[2]
SultainW, ChenC, ShahM. Real-world anomaly detection in surveillance videos[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos: IEEE Computer Society Press, 2018: 6479-6488.
[3]
TianY, PangG, ChenY, et al. Weakly-supervised video anomaly detection with robust temporal feature magnitude learning[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. New York: IEEE, 2021: 4975-4986.
[4]
ChenY, LiuZ, ZhangB, et al. MGFN: magnitude-contrastive glance-and-focus network for weakly-supervised video anomaly detection[J]. Proceedings of the AAAI Conference on ArtificialIntelligence, 2023, 37(1): 387-395.
[5]
ChoM A, KimM, HwangS, et al. Look around for anomalies: weakly-supervised anomaly detection via context-motion relational learning[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 12137-12146.
[6]
LiG, CaiG, ZengX, et al. Scale-aware spatio-temporal relation learning for video anomaly detection[C] // European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 333-350.
ZhangYibo, ZhangYali, ShuaiBoyu. Research on night anomal vehicle detection based on FaFPN[J]. Science Technology and Engineering, 2024, 24(27): 11869-11877.
[9]
WuJ C, HsiehH Y, ChenD J, et al. Self-supervised sparse representation for video anomaly detection[C] // European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 729-745.
[10]
PuY, WuX, YangL, et al. Learning prompt-enhanced context features for weakly-supervised video anomaly detection[J]. IEEE Transactions on Image Processing, 2024, 33: 4923-4936.
ZhangLin, ChenZhaobo, MaXiaoxuan, et al. Review of unsupervised and weakly supervised video anomaly detection methods[J]. Science Technology and Engineering, 2024, 24(19): 7941-7955.
ZhaoYuandi, HuYixin, TangShengjia, et al. Data anomaly detection and cleaning method based on flight trajectory prediction[J]. Science Technology and Engineering, 2025, 25 (26): 11398-11406.
DingChenke, ZhengSheng, HeJunjie, et al. Research on anomaly detection method for main pump in nuclear power plants based on dynamic multivariate state estimation[J]. Science Technology and Engineering, 2025, 25(27): 11614-11621.
ZhouGuangrong, YangSenquan, ZhengSheng, et al. Research on anomaly detection of steam generator water level based on NARX[J]. Science Technology and Engineering, 2024, 24(34): 14672-14678.
[19]
ChenY, DaiX, LiuM, et al. Dynamic convolution: attention over convolution kernels[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 11030-11039.
[20]
YangZ, ZhangT, BozchalooiI S, et al. Memory-augmented generative adversarial networks for anomaly detection[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 33(6): 2324-2334.
[21]
LuoW, ZhangT, YangW, et al. Action unit memory network for weakly supervised temporal action localization[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2021: 9969-9979.
[22]
ZhouY, YangP, QuY, et al. AnoOnly: Semi-supervised anomaly detection with the only loss on anomalies[J]. Expert Systems with Applications, 2025, 262: 125597.
[23]
WuP, LiuJ, ShiY, et al. Not only look, but also listen: learning multimodal violence detection under weak supervision[C]// Proceedings of the European Conference on Computer Vision (ECCV). Glasgow: Springer, 2020: 322-339.
[24]
ZhouH, YuJ, YangW. Dual memory units with uncertainty regulation for weakly supervised video anomaly detection[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2023, 37(3): 3769-3777.
[25]
TanW, YaoQ, LiuJ. Overlooked video classification in weakly supervised video anomaly detection[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. New York: IEEE, 2024: 202-210.
[26]
ZhangC, LiG, QiY, et al. Exploiting completeness and uncertainty of pseudo labels for weakly supervised video anomaly detection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 16271-16280.
[27]
LiS, LiuF, JiaoL. Self-training multi-sequence learning with transformer for weakly supervised video anomaly detection[J]. Proceedings of the AAAI Conference on Artificial Intelligence. 2022, 36(2): 1395-1403.
[28]
KarimH, DoshiK, YilmazY. Real-time weakly supervised video anomaly detection[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. New York: IEEE, 2024: 6848-6856.
[29]
WuP, ZhouX, PangG, et al. Open-vocabulary video anomaly detection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2024: 18297-18307.
[30]
WuP, LiuJ. Learning causal temporal relation and feature discrimination for anomaly detection[J]. IEEE Transactions on Image Processing, 2021, 30: 3513-3527.
[31]
ParkS, KimH, KimM, et al. Normality guided multiple instance learning for weakly supervised video anomaly detection[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. New York: IEEE, 2023: 2665-2674.
[32]
XuX, ChenC, LiangY, et al. SST: multi-scale hybrid mamba-transformer experts for long-short range time series forecasting[J]. arXiv Preprint, arXiv: 2404. 14757, 2024.
[33]
KangB, MoonS, ChoY, et al. Metaseg: Metaformer-based global contexts-aware network for efficient semantic segmentation[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. New York: IEEE, 2024: 434-443.
[34]
SalehiA, BalasubramanianM. DDCNet: deep dilated convolutional neural network for dense prediction[J]. Neurocomputing, 2023, 523: 116-129.
[35]
ZhouY, QuY, XuX, et al. Batchnorm-based weakly supervised video anomaly detection[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2024, 34(12): 13642-13654.
[36]
ChenW, MaK T, YewZ J, et al. TEVAD: improved video anomaly detection with captions[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 5549-5559.