A framework for dynamic restructuring of semantic video analysis systems based on learning attention control
作者：
Highlights：
• A framework is proposed to restructure an initial video analysis system dynamically.
• The initial structure of the system must be hierarchical containing some units.
• The framework imposes a learning-based dynamic feature selection method on each unit.
• The system learns how to direct attention to the informative units to achieve a goal.
• The framework is tested for goal and card event detection in broadcast soccer videos.
摘要
•A framework is proposed to restructure an initial video analysis system dynamically.•The initial structure of the system must be hierarchical containing some units.•The framework imposes a learning-based dynamic feature selection method on each unit.•The system learns how to direct attention to the informative units to achieve a goal.•The framework is tested for goal and card event detection in broadcast soccer videos.
论文关键词：Attention control,Broadcast soccer video,Event detection,Q-learning,Semantic video analysis
论文评审过程：Received 5 January 2015, Revised 8 July 2015, Accepted 10 July 2015, Available online 15 August 2015, Version of Record 9 September 2016.
论文官网地址：https://doi.org/10.1016/j.imavis.2015.07.004
