Hypergraph video pedestrian re-identification based on posture structure relationship and action constraints
作者：
Highlights：
• Model the relationship between the joint point regions based on the graph convolutional network, and obtain the visual and spatial structural features of the joint point regions.
• Based on the significant action information, the relationship between the joint point regions is constrained and the action hypergraph is constructed.
• Propose a loss function based on regional saliency score to metric the similarity of the saliency probability distribution of joint points in hypergraph space.
摘要
•Model the relationship between the joint point regions based on the graph convolutional network, and obtain the visual and spatial structural features of the joint point regions.•Based on the significant action information, the relationship between the joint point regions is constrained and the action hypergraph is constructed.•Propose a loss function based on regional saliency score to metric the similarity of the saliency probability distribution of joint points in hypergraph space.
论文关键词：Pedestrian re-identification,Structural relationship,Action hypergraph,Saliency score
论文评审过程：Received 12 March 2020, Revised 10 July 2020, Accepted 7 October 2020, Available online 8 October 2020, Version of Record 15 October 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2020.107688
