Human action recognition based on action relevance weighted encoding
作者：
Highlights：
• A new descriptor is proposed to highlight the discriminative trajectories in videos.
• A novel method to compute saliency map based on optical flow is proposed.
• Trajectory action relevance is defined to capture the importance of each trajectory.
• Frame action relevance is defined to describe the importance of each frame.
• A new video representation for human action recognition is proposed.
摘要
•A new descriptor is proposed to highlight the discriminative trajectories in videos.•A novel method to compute saliency map based on optical flow is proposed.•Trajectory action relevance is defined to capture the importance of each trajectory.•Frame action relevance is defined to describe the importance of each frame.•A new video representation for human action recognition is proposed.
论文关键词：Action recognition,Action relevance,Saliency map,Weighted encoding
论文评审过程：Received 12 February 2019, Revised 10 September 2019, Accepted 11 September 2019, Available online 13 September 2019, Version of Record 24 September 2019.
论文官网地址：https://doi.org/10.1016/j.image.2019.115640
