Online temporal classification of human action using action inference graph

作者：

Highlights：

• Traditional deep methods recognize action in a video sequence by averaging the results of all the clips/frames in the video sequence.

• Averaging over clips does not preserve the actual temporal order of motion and reduces the classification accuracy.

• We propose a new online temporal classification model (OTCM) for online action recognition to address the above mentioned limitation.

• We also propose a new action inference graph(AIG), which forms an important component of the proposed OTCM.

• The proposed OTCM requires much fewer frames to recognize an action. In some video sequences, it requires less than 10% of the video sequences.

摘要

•Traditional deep methods recognize action in a video sequence by averaging the results of all the clips/frames in the video sequence.•Averaging over clips does not preserve the actual temporal order of motion and reduces the classification accuracy.•We propose a new online temporal classification model (OTCM) for online action recognition to address the above mentioned limitation.•We also propose a new action inference graph(AIG), which forms an important component of the proposed OTCM.•The proposed OTCM requires much fewer frames to recognize an action. In some video sequences, it requires less than 10% of the video sequences.

论文关键词：Online temporal classification,Action recognition,Action detection,Action inference graph

论文评审过程：Received 30 September 2021, Revised 26 June 2022, Accepted 10 August 2022, Available online 11 August 2022, Version of Record 18 August 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108972