Dynamic mode decomposition via convolutional autoencoders for dynamics modeling in videos

作者：

Highlights：

• Extended dynamic mode decomposition for foreground modeling and video classification.

• Accurate extraction of foreground moving objects in videos.

• Background extraction by splitting low temporal frequency dynamic mode decomposition (DMD) modes.

• DMD modes of latent vectors are used as features for video classification.

摘要

•Extended dynamic mode decomposition for foreground modeling and video classification.•Accurate extraction of foreground moving objects in videos.•Background extraction by splitting low temporal frequency dynamic mode decomposition (DMD) modes.•DMD modes of latent vectors are used as features for video classification.

论文关键词：Dynamical systems,Dynamic mode decomposition,Nonlinear dynamics,Foreground modeling,Background modeling,Video classification

论文评审过程：Received 24 February 2021, Revised 29 November 2021, Accepted 27 December 2021, Available online 4 January 2022, Version of Record 20 January 2022.

论文官网地址：https://doi.org/10.1016/j.cviu.2021.103355