《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (5): 1452-1457.DOI: 10.11772/j.issn.1001-9081.2023050748

• 2023年中国计算机学会人工智能会议(CCFAI 2023) • 上一篇    

基于三元中心引导的弱监督视频异常检测

朱子蒙1, 李志新2, 郇战2(), 陈瑛2, 梁久祯1   

  1. 1.常州大学 计算机与人工智能学院、阿里云大数据学院、软件学院, 江苏 常州 213164
    2.常州大学 微电子与控制工程学院, 江苏 常州 213164
  • 收稿日期:2023-05-16 修回日期:2023-06-19 接受日期:2023-07-04 发布日期:2023-08-01 出版日期:2024-05-10
  • 通讯作者: 郇战
  • 作者简介:朱子蒙(1999—),男,甘肃兰州人,硕士研究生,CCF会员,主要研究方向:数据挖掘、异常检测、计算机视觉
    李志新(1993—),女,河南洛阳人,博士,CCF会员,主要研究方向:模式识别、计算成像和相位提取
    陈瑛(1987—),女,江苏常州人,讲师,博士,CCF会员,主要研究方向:数据挖掘、模式识别、物联网智能传感器
    梁久祯(1968—),男,山东泰安人,教授,博士生导师,博士,CCF会员,主要研究方向:计算机视觉、数据挖掘、模式识别、深度学习。
    第一联系人:郇战(1969—),男,陕西咸阳人,教授,博士生导师,博士,CCF会员,主要研究方向:数据挖掘、物联网智能传感器、智能感知
  • 基金资助:
    国家自然科学基金资助项目(62201093)

Weakly supervised video anomaly detection based on triplet-centered guidance

Zimeng ZHU1, Zhixin LI2, Zhan HUAN2(), Ying CHEN2, Jiuzhen LIANG1   

  1. 1.School of Computer Science and Artificial Intelligence,Aliyun School of Big Data,School of Software,Changzhou University,Changzhou Jiangsu 213164,China
    2.School of Microelectronics and Control Engineering,Changzhou University,Changzhou Jiangsu 213164,China
  • Received:2023-05-16 Revised:2023-06-19 Accepted:2023-07-04 Online:2023-08-01 Published:2024-05-10
  • Contact: Zhan HUAN
  • About author:ZHU Zimeng, born in 1999, M. S. candidate. His research interests include data mining, anomaly detection, computer vision.
    LI Zhixing, born in 1993, Ph. D. Her research interests include pattern recognition, computational imaging and phase retrieval.
    CHEN Ying, born in 1987, Ph. D., lecturer. Her research interests include data mining, pattern recognition, IoT smart sensor.
    LIANG Jiuzhen, born in 1968, Ph. D., professor. His research interests include computer vision, data mining, pattern recognition, deep learning.
  • Supported by:
    National Natural Science Foundation of China(62201093)

摘要:

针对监控视频异常的复杂多样性和短时持续性,引入弱监督视频异常检测方法,旨在仅使用视频级别的标签进行异常检测,并提出了基于变分自编码器(VAE)与长短期记忆(LSTM)网络的异常回归网络VLARNet作为异常检测框架,以捕获时序数据中的时间依赖关系、去除冗余信息,保留数据的关键信息。该框架将异常检测视为回归问题,为学习检测特征,设计了异常分数回归的三元中心损失(TCLASR),与动态多实例学习损失(DMIL)相结合以进一步提高特征的区分能力。DMIL能够扩大异常实例与正常实例之间的类间距离,但同时也扩大了类内距离,而TCLASR可使来自同类的实例与类中心的距离更接近,与不同类中心的距离更远。对VLARNet在ShanghaiTech与CUHK Avenue数据集上进行了综合实验。实验结果表明,VLARNet能够有效利用视频数据的各种信息,在两个数据集上获得的受试者工作特征曲线下面积(AUC)分别为94.64%和93.00%,明显优于对比算法。

关键词: 异常检测, 弱监督学习, 多实例学习, 中心损失, 受试者工作特征曲线下面积

Abstract:

In view of the complex diversity and short time persistence of surveillance video anomaly, a weakly supervised video abnormal detection method was introduced to detect anomalies by only using video-level tags, and an anomaly regression network VLARNet based on Variational AutoEncoder (VAE) and Long Short-Term Memory (LSTM) network was proposed as an anomaly detection framework to effectively capture the temporal dependencies in time series data, eliminate redundant information and retain key information in the data. Anomaly detection was considered as a regression problem by VLARNet. To learn detection features, a Triplet-Centered Loss for Anomaly Score Regression (TCLASR) was designed and combined with Dynamic Multiple Instance Learning loss (DMIL) to further improve the discrimination ability of features. The DMIL widened the inter-class distance between abnormal instances and normal instances, but it also widened the intra-class distance. The TCLASR made the distances between the instances in the same class and the center closer and the distances between instances in different classes and the center farther. The proposed VLARNet was comprehensively tested on ShanghaiTech and CUHK Avenue datasets. Experimental results show that VLARNet can effectively utilize various information in video data, achieving Area Under receiver operating characteristic Curve (AUC) of 94.64% and 93.00% respectively on the two datasets, which is significantly better than those of the comparison algorithms.

Key words: anomaly detection, weakly supervised learning, multiple instance learning, center loss, Area Under receiver operating characteristic Curve (AUC)

中图分类号: