Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Oct 14, 2025

Design and Analysis for Fall Detection System Simplification

Design and Analysis for Fall Detection System Simplification

Published on: April 6, 2020

Integrated Multiscale Appearance Features and Motion Information Prediction Network for Anomaly Detection.

Ting Liu¹, Chengqing Zhang^1,2, Liming Wang¹

¹State Key Lab for Electronic Testing Technology, North University of China, Taiyuan 030051, China.

Computational Intelligence and Neuroscience

|November 1, 2021

Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Scaling-basis chirplet extracting transform and its application in bearing fault diagnosis.

PloS one·2025

Same author

Local maximum synchrosqueezes form scaling-basis chirplet transform.

PloS one·2022

Same author

Spatio-temporal prediction and reconstruction network for video anomaly detection.

PloS one·2022

Same author

Underwater acoustic target recognition method based on a joint neural network.

PloS one·2022

Same author

Interaction of arene ruthenium(II) complexes [(η<sup>6</sup>-C<sub>6</sub>H<sub>6</sub>)Ru(L)Cl]PF<sub>6</sub> (L = o-fpip and p-fpip) with the RNA triplex poly(U)*poly(A)•poly(U).

Journal of inorganic biochemistry·2022

Same author

3D M-Net: Object-Specific 3D Segmentation Network Based on a Single Projection.

Computational intelligence and neuroscience·2021

This study introduces a novel video prediction network for enhanced anomaly detection in surveillance. The method improves accuracy by integrating multi-scale appearance and motion features, outperforming existing techniques.

Area of Science:

Computer Vision
Artificial Intelligence
Machine Learning

Background:

Video prediction algorithms are crucial for anomaly detection in smart city surveillance.
Current methods often fail to capture spatiotemporal information due to reliance on single-scale features and lack of temporal continuity.

Purpose of the Study:

To propose a novel prediction network for improving anomaly detection accuracy in video surveillance.
To address the limitations of existing methods in capturing multi-scale appearance and temporal motion information.

Main Methods:

Utilized a hybrid dilated convolution (HDC) module for multi-scale appearance feature extraction.
Employed a deeper bidirectional convolutional long short-term memory (DB-ConvLSTM) module to retain motion continuity.

Related Experiment Videos

Last Updated: Oct 14, 2025

Design and Analysis for Fall Detection System Simplification

Design and Analysis for Fall Detection System Simplification

Published on: April 6, 2020

Replaced optical flow loss with RGB difference loss for efficient temporal constraint.

Main Results:

The proposed network effectively extracts detailed appearance features using varied receptive fields.
The DB-ConvLSTM module successfully preserves motion information across video frames.
RGB difference loss significantly reduces computation time compared to optical flow extraction.

Conclusions:

The novel prediction network demonstrates superior performance in anomaly detection across diverse surveillance scenarios.
The integration of HDC and DB-ConvLSTM modules enhances spatiotemporal information capture for accurate abnormality detection.
The RGB difference loss offers an efficient alternative for temporal constraints, improving overall system performance.