Towards zero shot learning of geometry of motion streams and its application to anomaly recognition
作者：
Highlights：
• Geometry of temporal-derivatives contains discriminatory features.
• Late fusion of temporal-derivatives declines recognition performance.
• Local anomalies have lower recognition rates than global.
• Variation in temporal scales requires spatial–temporal fusion.
• Deep flow has stronger priors than foreground segmentation.
摘要
•Geometry of temporal-derivatives contains discriminatory features.•Late fusion of temporal-derivatives declines recognition performance.•Local anomalies have lower recognition rates than global.•Variation in temporal scales requires spatial–temporal fusion.•Deep flow has stronger priors than foreground segmentation.
论文关键词：Unsupervised learning,Temporal derivatives,Multilinear algebra,Visual anomaly recognition
论文评审过程：Received 8 October 2019, Revised 23 January 2021, Accepted 14 March 2021, Available online 18 March 2021, Version of Record 7 April 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.114916
