Two video analysis applications using foreground/background segmentation

Abstract

Probably the most frequently solved problem when videos are analyzed is segmenting a foreground object from its background in an image. After some regions in an image are detected as the foreground objects, some features are extracted that describe the segmented regions. These features together with the domain knowledge are often enough to extract the needed high-level semantics from the video material. In this paper we present two automatic systems for video analysis and indexing. In both systems the segmentation of the foreground objects is the basic processing step. The extracted features are then used to solve the problem. One system analyses traffic videos, the other tennis games.