Event Detection And Tracking In News Video

Due to advances in telecommunication and multimedia technology, the amount of information is increasing as never before. However, the relative information is distributed about different lactions and piroids, it is important to research the technologies to automatically detect, classify and summarize the relative information. In this paper, we studied the technologies of structurizing and sematic extraxtion of video, and proposed the methods for these issues. And on basis of the above work, we studied the news story segmentation, news topic detection and tracking in news video, and also proposed the mothed for news event detection.Video structurizeing is the base of the further processes. From top to bottom increasing in granularity, videos can be represented by a hierarchical structure consisting of 4 levels: video, scene, shot and frame, in while the shot and scene segmentation is the key issues. In this paper, a mothed based on SVM for shot segmentation is proposed as well a statistical learning mothed for scene segmentation. For shot segmentation, a variance projection function is introduced, and by using this variance projection function, the distance between the video frames is defined, then a statistical learning method based on the support vector machine is devised to find the boundary of shot. For scene detection, video structure is first analyzed by combining spatial-temporal analysis and statistical learning; scenes are then detected based on unsupervised statistical learning.News story segmentation is a prior condition for news analysis. We propose an approach for news story segmentation based on data association mining in this paper. First, we transform video into data stream which consists of involved events; then define the support and confidence, and explore the patterns corresponding to news from the data stream using association mining techniques; and then search the section which matches the patterns in the data stream to find the pattern for news story.Based on the above analysis of the news video, we present a probabilistic learning approach to model video news story for topic detection and tracking, In this approach, both content and time information of a news video is utilized to transcribe the news story into terms, which are divided into classes by their semantics. Then a probabilistic model, composed of sub-models corresponding to the semantic classes respectively, is proposed for new event detection and event trospective detection.