基于动态贝叶斯网络的体育视频分析  

Sports Video Analysis Based on Dynamic Bayes Net

作　　者：刘云[1] 刘雪英[1] 黄超[1] 

机构地区：[1]青岛科技大学信息科学技术学院,青岛市266061

出　　处：《微计算机信息》2010年第21期9-10,61,共3页Control & Automation

基　　金：基金申请人:刘云;项目名称:基于规则和机器学习的视频标注新算法研究;基金颁发部门:山东省自然科学基金委(Y2008G09)

摘　　要：视频分析的核心是对语义事件及其相互关系的分析,其主要任务是解决底层特征和高层语义之间的"语义鸿沟"。文中研究了基于动态贝叶斯网络的体育视频分析。首先,将视频流分解为一个个组成语义事件的底层基元,然后通过Kalman滤波原理提取图像特征,并运用EM算法完成DBN参数学习。实验结果证明了本文方法的有效性。The core of the video analysis is analyzing the relationship between the semantic events and its main task is to solve the ＂semantic gap＂. This paper researches on the use of DBN to analyze the sports video. Firstly, the video stream is decomposed into some bottom primitives, which comprise semantic events. And then kalman filter is used to extract the features, and the EM algorithm is used to learn DBN parameters. The experiment results verify that this algorithm is very effective.

关 键 词：DBN 视频分析 KALMAN滤波 EM算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]