Memory-augmented deep autoencoder model for pedestrian abnormal behavior detection in video surveillance

Jingbo Sun; Jie Ji

doi:10.3788/IRLA20210680

Journals >Infrared and Laser Engineering >Volume 51 >Issue 6 >Page 20210680 > Article

Infrared and Laser Engineering
Vol. 51, Issue 6, 20210680 (2022)

Memory-augmented deep autoencoder model for pedestrian abnormal behavior detection in video surveillance

Jingbo Sun and Jie Ji

Author Affiliations

School of Mathematics and Computer Application Technology, Jining University, Qufu 273155, China

show less

DOI: 10.3788/IRLA20210680 Cite this Article

Jingbo Sun, Jie Ji. Memory-augmented deep autoencoder model for pedestrian abnormal behavior detection in video surveillance[J]. Infrared and Laser Engineering, 2022, 51(6): 20210680 Copy Citation Text

show less

Fig. 1. The flow chart of Memory AE based anomaly detection method

Download full size | View in the Article

Fig. 2. Examples of the detection results

Download full size | View in the Article

Method	Avenue	ShanghaiTech
MPPCA+ SF ^[17]	56.2%	-
MDT^[18]	77.4%	-
Conv-AE ^[8]	80.0%	60.9%
Conv3D-AE^[19]	80.9%	-
Stacked RNN^[20]	81.7%	68.0%
ConvLSTM-AE^[21]	77.0%	-
MemNormality^[22]	88.5%	70.5%
ClusterAE^[23]	86.0%	73.3%
AbnormalGAN^[24]	-	72.4%
Pred+Recon^[25]	85.1%	73.0%
Proposed method	85.7%	75.3%

Table 1. Comparison with the state of the art methods in terms of AUC%

View in the Article

Size of memory module	500	1000	1500	2000	2500
Result	78.2%	85.7%	85.3%	85.7%	85.8%

Table 2. The influence of the number of memory module size on the experimental results of the Avenue data set (frame-level AUC%)

Jingbo Sun, Jie Ji. Memory-augmented deep autoencoder model for pedestrian abnormal behavior detection in video surveillance[J]. Infrared and Laser Engineering, 2022, 51(6): 20210680

Download Citation

Tools

Save the article for my favorites

Paper Information