Semantic human activity recognition: A literature review
作者：
Highlights：
• We have conducted a detailed review on recent action recognition frameworks based on semantic information for the first time.
• A semantic space is introduced that mainly includes pose, poselet, object/scene context, and attributes.
• We discuss that semantic descriptions capture meaningful information and are robust to visual variations.
• We show that semantic methods outperform non-semantic methods in most cases with comparing their performances.
• We present some directions to motivate future researchers to devote more attention to semantic information.
摘要
Highlights•We have conducted a detailed review on recent action recognition frameworks based on semantic information for the first time.•A semantic space is introduced that mainly includes pose, poselet, object/scene context, and attributes.•We discuss that semantic descriptions capture meaningful information and are robust to visual variations.•We show that semantic methods outperform non-semantic methods in most cases with comparing their performances.•We present some directions to motivate future researchers to devote more attention to semantic information.
论文关键词：Human activity recognition,Pose,Poselet,Attribute,Human-object interaction,Scene,Survey
论文评审过程：Received 2 October 2014, Revised 7 February 2015, Accepted 6 March 2015, Available online 14 March 2015.
论文官网地址：https://doi.org/10.1016/j.patcog.2015.03.006
