Attention-based Residual Autoencoder for Video Anomaly Detection

Abstract

Automatic anomaly detection is a crucial task in video surveillance system intensively used for public safety and others. The present system adopts a spatial branch and a temporal branch in a unified network that exploits both spatial and temporal information effectively. The network has a residual autoencoder architecture, consisting of a Resnet-based encoder and a multi-stage channel attention-based decoder, trained in an unsupervised manner. The residual temporal shift is used for exploiting the temporal feature, whereas the contextual dependency is extracted by channel attention modules. System performance is evaluated using three standard benchmark datasets. Result suggests that our network outperforms the state-of-the-art methods, achieving 97.4% for USCD Ped2, 86.7% for CUHK Avenue, and 73.6% for ShanghaiTech dataset in term of Area Under Curve, respectively.

Method

Video

Anomaly Score

Ped2

Avenue

ShanghaiTech

Related Works

• HSTforU: Anomaly Detection in Aerial and Ground-based Videos with Hierarchical Spatio-Temporal Transformer for U-net
• MoGuP:Motion-guided Prediction for Video Anomaly Detection

BibTeX

@article{le2023attention,
    author    = {Le, Viet-Tuan and Kim, Yong-Guk},
    title     = {Attention-based Residual Autoencoder for Video Anomaly Detection},
    journal   = {Applied Intelligence},
    volume    = {53},
    number    = {3},
    pages     = {3240--3254},
    year      = {2023},
    publisher = {Springer}
}

Acknowledgements

This work was supported by the Institute of Information & communications Technology Planning & Evaluation (IITP), grant funded by the Korea government (MSIT) (No.2019-0-00231).