Extraction of Semantic Dynamic Content from Videos with Probabilistic Motion Models

Abstract

Abstract. The exploitation of video data requires to extract information at a rather semantic level, and then, methods able to infer "concepts" from low-level video features. We adopt a statistical approach and we focus on motion information. Because of the diversity of dynamic video content (even for a given type of events), we have to design appropriate motion models and learn them from videos. We have defined original and parsimonious probabilistic motion models, both for the dominant image motion (camera motion) and the residual image motion (scene motion). These models are learnt off-line. Motion measurements include affine motion models to capture the camera motion, and local motion features for scene motion. The two-step event detection scheme consists in pre-selecting the video segments of potential interest, and then in recognizing the specified events among the pre-selected segments, the recognition being stated as a classification problem. We report accurate results on several sports videos.

Multiscale NeighborhoodWise Decision Fusion for Redundancy Detection in Image Pairs

“…There has been a substantial amount of work to handle changes in an image pair [3,99,2,86,66,80,1,49]. For a recent survey, see [83].…”

Recognition of Dynamic Video Contents With Global Probabilistic Models of Visual Motion

“…We have designed a two-step event detection method to restrict the recognition issue to a limited and pertinent set of classes, since probabilistic motion models have to be learned for each class of dynamic content to be recognized. A preliminary version of this work was described in [32]. The two-step framework allows us to simplify the learning stage, to save computation time and to make the overall detection more robust and efficient.…”

2D Motion Description and Contextual Motion Analysis: Issues and New Models

“…The motion information is captured through low-level motion measurements so that it can be efficiently and reliabily computed in any video whatever its genre and its content. Our approach (joint work with Gwénaëlle Piriou and Jian-Feng Yao [40]) consists in modeling separately the camera motion (i.e., the dominant image motion) and the scene motion (i.e., the residual image motion) in a sequence, since these two sources of motion bring important and complementary information. The dominant image motion is represented by a deterministic 2D affine motion model (which is a usual choice):…”

“…The number of components of the mixture is determined with the Integrated Completed Likelihood criterion (ICL) and their parameters are estimated using the ExpectationMaximization (EM) algorithm [40].…”

“…Therefore, we consider a specific mixture model to represent the distribution of the local residual motion measurements within a video segment with density [40]:…”