Abstract: The semantic clues embedded in video motion features are plenteous.Concise representation and effective extraction method for motion feature are the key techniques in video semantic analysis processing.According to the characteristic of video semantic analysis，the motion features are classified to three categories.Then，representation and extraction for each category are given.The experiments indicate the representation and extraction method proposed can work effectively.Finally，a prototype video semantic analysis system based on motion features is realized.

摘要： 视频运动特征蕴含丰富的语义信息，运动特征的简洁表征方式和高效抽取方法研究是视频语义分析的关键技术之一。针对视频语义分析的特点，将运动特征分为3类，分别对各类运动特征进行表征和抽取。相关抽取实验证明此方法可有效抽取语义分析所需的运动特征，同时在运动特征抽取的基础上实现了基于运动的视频语义分析原型系统。