Earth mover's prototypes: A convex learning approach for discovering activity patterns in dynamic scenes

摘要: We present a novel approach for automatically discovering spatio-temporal patterns in complex dynamic scenes. Similarly to recent non-object centric methods, we use low level visual cues detect atomic activities and then construct clip histograms. Differently from previous works, formulate the task of high activity as prototype learning problem where correlation among is explicitly taken into account when grouping Interestingly at core our there convex optimization which allows us efficiently extract multiple levels detail. The effectiveness method demonstrated on publicly available datasets.