Extraction of semantic dynamic content from videos with probabilistic motion models

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. © Springer-Verlag 2004.