• Infrared and Laser Engineering
  • Vol. 51, Issue 6, 20210547 (2022)
Youkun Zhong1 and Haining Mo2、*
Author Affiliations
  • 1Physics and Mechanical & Electrical Engineering School, Hechi University, Yizhou 546300, China
  • 2HTC VIVEDU School of Technology, Guangxi University of Science and Technology, Liuzhou 545006, China
  • show less
    DOI: 10.3788/IRLA20210547 Cite this Article
    Youkun Zhong, Haining Mo. A video anomaly detection method based on deep autoencoding Gaussian mixture model[J]. Infrared and Laser Engineering, 2022, 51(6): 20210547 Copy Citation Text show less
    Flow chart of abnormal event detection method based on DAGMM
    Fig. 1. Flow chart of abnormal event detection method based on DAGMM
    Examples of the detection results
    Fig. 2. Examples of the detection results
    AttributesUCSD Ped2ShanghaiTech
    Frames4560317398
    SceneSingleMulti
    LabelsSpatial & TemporalSpatial & Temporal
    Resolution360×240856×480
    AnomaliesBiker, cart, etcChasing, brawling sudden motion, etc
    Table 1. Overview of benchmark datasets
    MethodUCSD Ped2ShanghaiTech
    MPPCA[3]69.3%-
    MDT[4]82.9%-
    MT-FRCN[5]92.2%-
    Conv2D-AE[10]85.0%60.9%
    Conv3D-AE[10]91.2%-
    ConvLSTM-AE[20]88.1%-
    StackRNN[21]92.2%68.0%
    Baseline[18]95.4%72.8%
    Proposed method95.7%72.9%
    Table 2. Comparison with the state of the art methods in terms of AUC%
    $ K $AUC%
    292.3%
    494.5%
    8951%
    1695.7%
    3295.6%
    6495.7%
    Table 3. Influence of the number of Gaussian mixture components number K on the experimental results of the UCSD Ped2 data set (frame-level AUC%)
    Youkun Zhong, Haining Mo. A video anomaly detection method based on deep autoencoding Gaussian mixture model[J]. Infrared and Laser Engineering, 2022, 51(6): 20210547
    Download Citation