Abstract
The exploitation of semantic information in computer vision problems can be difficult because of the large difference in representations, levels of knowledge and abstract episodes. Traditional image/video understanding, indexing is formulated in terms of low-level features describing image/video structure and intensity, while high-level knowledge such as common sense and human perceptual knowledge are encoded in abstract, non-geometric representations. In this thesis, we attempt to bridge this gap through the integration of image/video analysis algorithms with multi-level Bayesian Brief Network (BBN), a large semantic network that explicitly links related words in a hierarchical structure. Our problem domain is the understanding of sports programs, as this provides both linguistic information in site information and special efficacy of view. Visual detection algorithms such as scene change detection, moving object segmentation and morphology analysis combined with low–level feature extract algorithms are applied to the video to extract the basic object/background information as the input to the Bayesian Belief Network. Our video understanding system is designated for the baseball game video program, in which the events may be the scenes of pitcher pitching, batter running, or the outfielder catching the ball, etc. Given a video in a specific domain, our system may extract the low-level evidences and then interpret the input video by high-level semantic.