Human action recognition by means of subtensor projections and dense trajectories

作者：

Highlights：

• We show that temporal templates preserve useful information for action recognition.

• We have assessed the optimal projection functions and feature descriptors.

• Using optimum parameters, they can overcome most of state of art techniques.

• Our templates combined with HOF, HOG and MBH improve the results of IDTs. We present several exhaustive tests to prove it.

摘要

•We show that temporal templates preserve useful information for action recognition.•We have assessed the optimal projection functions and feature descriptors.•Using optimum parameters, they can overcome most of state of art techniques.•Our templates combined with HOF, HOG and MBH improve the results of IDTs. We present several exhaustive tests to prove it.

论文关键词：Action recognition,Subtensors,Dense trajectories,Keypoint descriptors,Temporal template

论文评审过程：Received 20 September 2017, Revised 3 April 2018, Accepted 11 April 2018, Available online 12 April 2018, Version of Record 22 April 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.04.015