Fine-grained action recognition using dynamic kernels

作者：

Highlights：

• Action-independent Gaussian mixture model (AIGMM) is constructed to preserve local similarity among fine-grained actions.

• We propose an approach to handle variable-length patterns of fine-grained actions using the statistics of trained AIGMM.

• Efficacy of our approach is demonstrated on 4 fine-grained action datasets, namely, MERL, JIGSAWS, KSCGR, & MPII cooking2.

摘要

•Action-independent Gaussian mixture model (AIGMM) is constructed to preserve local similarity among fine-grained actions.•We propose an approach to handle variable-length patterns of fine-grained actions using the statistics of trained AIGMM.•Efficacy of our approach is demonstrated on 4 fine-grained action datasets, namely, MERL, JIGSAWS, KSCGR, & MPII cooking2.

论文关键词：Fine-grained action recognition,Spatio-temporal features,Gaussian mixture model,Dynamic kernels

论文评审过程：Received 16 April 2021, Revised 20 August 2021, Accepted 28 August 2021, Available online 30 August 2021, Version of Record 7 September 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108282