Robust action recognition using local motion and group sparsity
作者：
Highlights：
• Dense sampling based approaches dilute the information about motions of interest.
• The proposed algorithm identifies local motions of interest using group sparsity.
• We emphasize the local motion using multiple kernel method.
• It alleviates the effect of the contaminated descriptors.
• We demonstrate that the proposed method is robust for realistic video sequences.
摘要
Highlights•Dense sampling based approaches dilute the information about motions of interest.•The proposed algorithm identifies local motions of interest using group sparsity.•We emphasize the local motion using multiple kernel method.•It alleviates the effect of the contaminated descriptors.•We demonstrate that the proposed method is robust for realistic video sequences.
论文关键词：Action recognition,Motion descriptor,Sparse representation,Dynamic scene understanding
论文评审过程：Received 17 June 2013, Revised 4 December 2013, Accepted 5 December 2013, Available online 16 December 2013.
论文官网地址：https://doi.org/10.1016/j.patcog.2013.12.004
