Handcrafted localized phase features for human action recognition

作者：

Highlights：

• Proposal of phase correlation to model motion leading to specific human actions.

• Proposal of considering localized phase variation to extract features for representing human actions.

• Evaluation of the proposed model with commonly used HAR datasets.

• Outperforming currently state-of-the art Deep Learning methods on challenging datasets like Kinetics-400 and Kinetics-700.

摘要

•Proposal of phase correlation to model motion leading to specific human actions.•Proposal of considering localized phase variation to extract features for representing human actions.•Evaluation of the proposed model with commonly used HAR datasets.•Outperforming currently state-of-the art Deep Learning methods on challenging datasets like Kinetics-400 and Kinetics-700.

论文关键词：Motion analysis,Phase analysis,Human action recognition,Handcrafted features

论文评审过程：Received 31 January 2022, Revised 17 April 2022, Accepted 18 April 2022, Available online 25 April 2022, Version of Record 5 May 2022.

论文官网地址：https://doi.org/10.1016/j.imavis.2022.104465