Human action recognition with graph-based multiple-instance learning

作者：

Highlights：

• A motion compensation approach based on affine model is introduced.

• A density discontinuity detector is proposed to deal with over-segmentation.

• The bag of histograms is presented to develop the bag of features.

• A graph-based multiple-instance learning method is proposed to handle ambiguity.

摘要

•A motion compensation approach based on affine model is introduced.•A density discontinuity detector is proposed to deal with over-segmentation.•The bag of histograms is presented to develop the bag of features.•A graph-based multiple-instance learning method is proposed to handle ambiguity.

论文关键词：Action recognition,Dense trajectory,Motion compensation,Spectral embedding,Multiple-instance learning

论文评审过程：Received 26 April 2015, Revised 31 August 2015, Accepted 28 November 2015, Available online 3 December 2015, Version of Record 8 February 2016.

论文官网地址：https://doi.org/10.1016/j.patcog.2015.11.022