On an algorithm for human action recognition

作者：

Highlights：

• LMDI based interest point detection technique is proposed to detect non 3D corners.

• Overlapping split is adopted into Random Projection tree.

• Modified RP Tree efficiently handles the points lying near the boundary.

• Voting technique is well supported by vote counts in shape model.

摘要

•LMDI based interest point detection technique is proposed to detect non 3D corners.•Overlapping split is adopted into Random Projection tree.•Modified RP Tree efficiently handles the points lying near the boundary.•Voting technique is well supported by vote counts in shape model.

论文关键词：Activity recognition,Interest point,Random projection tree,Local feature,Hough voting

论文评审过程：Received 5 December 2017, Revised 21 June 2018, Accepted 10 August 2018, Available online 10 August 2018, Version of Record 25 August 2018.

论文官网地址：https://doi.org/10.1016/j.eswa.2018.08.014