The importance of high level representation in Action Recognition

Application Action recognition can be profitably used in several application fields, ranging from surveillance systems and ambient assisted living, with the possibility of identifying some events of interest occurring into a video (an aggression, a robbery, a tumble of an elder person, etc), to video retrieval, by allowing a human operator to easily retrieve specific types of actions performed by persons in a video footage. The recognition of an action is typically performed at two different layers: at a low level, a set of features (local or global) is analysed frame by frame; at a high level, the temporal information is explored for taking the final decision regarding the action to be recognized.