Automatically parsing and labelling video based on camera motion qualitative analysis

Structurized video representation and indexing based on the representation are the foundation of video retrieval based on content. A novel algorithm is presented of automatically parsing and labelling video based on camera motion qualitative analysis. Firstly, the motion patterns are analyzed of the spatio-temporal slices in a video; and then the change detection and qualitative recognition of camera motion are carried out by using motion structure tensor. Finally, the results of camera motion analysis are used to parse and label video according to the camera motion coherence. The algorithm is experimentally proved effective and produces satisfactory results.