摘要： 以网球视频为例，提出一种基于语义的体育视频场景分割方法。基于网球视频的先验知识设计一个具有6个语义场景的分类器，并根据各个场景的视觉特点提取球场地标线连接点、球场颜色、相机运动模式和人物等可感知特征作为特征。利用支持向量机技术对视频镜头进行语义分类，并给出一种利用聚类提取示例的主动学习算法。对大量网球视频进行实验，结果表明该方法能够得到比传统方法更好的效果。

关键词: 视频检索, 主动学习, 语义

Abstract: This paper presents techniques and results on semantic annotation of sport video with active learning. It takes tennis video as examples and defines 6 scenes containing semantic concept. The mid-level features such as court line, court color percentage, domain motion, camera move pattern etc are extracted through analyzing the visual difference of semantic scenes. The Support Vector Machine(SVM) is used to classify the shots. The proposed method employs a clustering based active learning scheme. It performs most representative samples selection as well as to avoid repeatedly labeling samples in the same cluster. Experimental results show encouraging result compared with the traditional methods.

Key words: video indexing, active learning, semantic

中图分类号: 

• TN911.73