Exploring trace transform for robust human action recognition

作者：

Highlights：

• We propose two novel feature extraction methods for Human Action Recognition (HAR).

• We examine for the first time, according to our knowledge, the capabilities of Trace transform for HAR.

• We introduce History Trace Templates (HTTs) and History Triple Features (HTFs) as a way to capture the spatiotemporal information of a human action.

• We employ different functionals of the Trace transform to calculate robust features and we introduce new features invariant to video caption variations.

• Experiments performed on noisy data without prior filtering, indicated great potentiality and very competitive results.

摘要

Highlights•We propose two novel feature extraction methods for Human Action Recognition (HAR).•We examine for the first time, according to our knowledge, the capabilities of Trace transform for HAR.•We introduce History Trace Templates (HTTs) and History Triple Features (HTFs) as a way to capture the spatiotemporal information of a human action.•We employ different functionals of the Trace transform to calculate robust features and we introduce new features invariant to video caption variations.•Experiments performed on noisy data without prior filtering, indicated great potentiality and very competitive results.

论文关键词：Human action recognition,Motion analysis,Action classification,Trace transform

论文评审过程：Received 27 March 2012, Revised 26 February 2013, Accepted 1 June 2013, Available online 15 June 2013.

论文官网地址：https://doi.org/10.1016/j.patcog.2013.06.006