Efficient access to information contained in video databases implies that a structured representation of the content of the video is built beforehand. We will address several important issues to be solved in that context: - temporal segmentation of the video into shots handling different kinds of transitions (cut, fade, dissolve, wipe) - camera motion characterization (static, panning, zooming, tracking) - extraction and tracking of relevant entities within a shot - scene activity characterization for a large range of motion types - creation of "annotated" iconic video summaries and index database - motion-based query by similarity. We have build an original and unified framework which mainly proceeds from the analysis of the apparent motion between successive images and of related information. It relies on statistical techniques, involving robust estimation of 2D parametric motion models, Hinkley's test, log-likelihood ratio tests, MRF modeling, statistical contextual labeling, temporal cooccurrences of local motion-related quantities. References:
Some of them are available by a click here.