基于统一特征模型的体育视频镜头转换检测  

Enhanced Sports Video Shot Transition Detection Based on a Unified Feature Model

作　　者：韩博[1] 胡一川[1] 吴伟国[2] 王贵锦[1] Takayuki Yoshigahara 

机构地区：[1]清华大学电子工程系,北京100084 [2]索尼中国研究院,北京100080 [3]索尼信息技术研究院

出　　处：《电视技术》2007年第8期76-79,共4页Video Engineering

基　　金：国家自然科学基金(60472028);教育部博士点基金(20040003015)

摘　　要：提出了一种统一特征模型,集成了2种新型的基于主颜色分割与运动矢量可信度分析的中层特征。这一模型根据主颜色分割结果自适应地采用不同特征和相应的分类界面,通过时域统计特征的使用在支持向量机中自适应地确定分类阈值。通过在较大的足球视频数据集上实验表明,本文算法对切变和渐变的检测都超过了已有算法,基本达到体育视频检索的应用要求。A unified feature model is proposed to solve the main problems encountered in this task. It integrates two novel middle level features, which are based on dominant color segmentation and motion vector reliability analysis. This model adaptively selects features and the corresponding decision plane based on the dominant color segmentation result; it adaptively sets the thresholds via using time domain statistical features as the inputs of support vector machines. Experiments on a relatively large soccer video dataset demonstrate that, for both cut detection and gradual transition detection, the scheme outperforms the existing algorithms and is quite competent for sports video retrieval applications.

关 键 词：镜头转换检测 统一特征模型 中层特征 体育视频分析 视频检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]