Abstract
The exploitation of semantic information from video is a nontrivial problem 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 Belief Network (BBN), and demonstrate how we can be effectively applied for fusing the evidence obtained from different video sources. Support vector tracking is applied for ball/shooter/basket tracking. SVM classifier and camera motion analysis combined with low–level feature extract algorithms are applied to extract mid-level features from the video which act as the input to the Bayesian Belief Network. We have proposed a novel video shot classification system based on low-level features extraction. Our semi-automatic semantic system is designated for the basketball game videos. Given the video shots of basketball game, our framework can identify four categories of shot event such as short shot, medium shot, long shot, free throw, and the score event. In the experiments, we demonstrate that our system may extract the low-level evidences and then interpret the high-level semantic of the video shot.