Opto-Electronic Engineering, Volume. 51, Issue 5, 240034(2024)

Improved spatio-temporal graph convolutional networks for video anomaly detection

Hongmin Zhang*... Dingding Yan and Qianqian Tian |Show fewer author(s)
Author Affiliations
  • School of Electrical and Electronic Engineering, Chongqing University of Technology, Chongqing 400054, China
  • show less
    Figures & Tables(13)
    Improved spatio-temporal graph convolutional network model framework
    Comparison between GCN module and CRF-GCN module. (a) GCN module; (b) CRF-GCN module
    Flowchart of mean-field inference for CRF-GCN
    Test results of UCSD Ped2 dataset. (a) Test003; (b) Test012
    Test results of ShanghaiTech dataset. (a) 04_0004; (b) 12_0173
    Test results of IITB-Corridor dataset. (a) Test000228; (b) Train000139 (Normal)
    Noised experiments. (a) AUC loss for training with noise-added data; (b) ACC loss for training with noise-added
    • Table 1. UCSD Ped2, ShanghaiTech and IITB-Corridor datasets

      View table
      View in Article

      Table 1. UCSD Ped2, ShanghaiTech and IITB-Corridor datasets

      数据集帧数年份标注分辨率异常类型
      UCSD Ped245602010Frame-level360×240骑自行车、小型车辆
      ShanghaiTech3173982016Frame-level480×856骑自行车、逃票、打架
      IITB-Corridor4835662020Frame-level1920×1080抗议、打斗、追逐等
    • Table 2. Comparison results of different methods on UCSD Ped2 dataset

      View table
      View in Article

      Table 2. Comparison results of different methods on UCSD Ped2 dataset

      监督方式对比方法特征提取方式AUC/%准确率/%
      无监督方式Hasan的方法[28]-90.089.5
      Gong的方法[29]-94.1-
      Yu的方法[30]-97.395.6
      Taghinezhad的方法[31]Encoder97.6-
      弱监督方式GCN-Anomaly[27]TSN93.290.3
      Sultani的方法[7]I3D92.3-
      RTFM[32]TSN96.5-
      Chen的方法[33]C3D97.496.1
      Wang的方法[34]Encoder97.793.4
      本文方法C3D97.796.5
    • Table 3. Comparison results of different methods on ShanghaiTech dataset

      View table
      View in Article

      Table 3. Comparison results of different methods on ShanghaiTech dataset

      监督方式对比方法特征提取方式AUC/%准确率/%
      无监督方式Hasan的方法[28]-60.860.1
      Gong的方法[29]-71.2-
      Yu的方法[30]-74.472.6
      Tur的方法[35]3D-ResNet1876.1-
      弱监督方式GCN-Anomaly[27]TSN84.482.6
      Sultani的方法[7]I3D86.3-
      Zhou的方法[12]I3D89.8-
      Acsintoae的方法[36]-83.786.1
      Wang的方法[34]Encoder71.382.6
      本文方法C3D90.488.6
    • Table 4. Comparison results of different methods on IITB-Corridor dataset

      View table
      View in Article

      Table 4. Comparison results of different methods on IITB-Corridor dataset

      监督方式对比方法特征提取方式AUC/%
      无监督方式Zeng的方法[37]-73.9
      弱监督方式Li的方法[38]C3D72.2
      Cao的方法[39CVAE73.6
      Royston的方法[26]I3D67.1
      Majhi的方法[40]I3D84.1
      本文方法C3D86.0
    • Table 5. Comparison results of different methods on complexity

      View table
      View in Article

      Table 5. Comparison results of different methods on complexity

      分类对比方法MACs/GParams/M
      基于其他框架的方法Sultani的方法[7]154.2263.33
      Feng的方法[19]156.8634.75
      基于图卷积的方法GCN-Anomaly[27]154.2263.38
      Chen的方法[33]154.2363.90
      本文方法109.1419.90
    • Table 6. Results of ablation experiments

      View table
      View in Article

      Table 6. Results of ablation experiments

      时间依赖图空间相似图图融合方式CRFAUC/%准确率/%
      -96.696.2
      -97.196.1
      平均融合[29]89.286.9
      自适应时空融合96.194.2
      自适应时空融合97.796.5
    Tools

    Get Citation

    Copy Citation Text

    Hongmin Zhang, Dingding Yan, Qianqian Tian. Improved spatio-temporal graph convolutional networks for video anomaly detection[J]. Opto-Electronic Engineering, 2024, 51(5): 240034

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Article

    Received: Feb. 1, 2024

    Accepted: Apr. 10, 2024

    Published Online: Jul. 31, 2024

    The Author Email: Zhang Hongmin (张红民)

    DOI:10.12086/oee.2024.240034

    Topics