中文摘要：

本文对基于运动矢量的运动量的建模方法进行了修正，同时引入语速这种音频特征，在综合了镜头变换率和声音能量两个特征基础之上，提出了一种新的兴奋建模和视频摘要统一框架。在该框架之内，首先对视频的兴奋内容进行建模得到兴奋时间曲线，之后，依据曲线中的极大值和极小值提取关键帧和精彩片断两种形式的视频摘要。实验证明本文建模方法是有效的，提取的视频摘要能够有效表示视频的内容，且具有良好的面向用户性和自适应性。提取的关键帧的有效性达到78％，足球片断的精彩有效性和排序的有效性分别达到81％和82％，篮球片断的精彩有效性和排序的有效性分别达到75％和76％。

英文摘要：

In this paper, the modeling approach based on motion vectors was improved, and a new audio feature （ i. e. speech rate） was introduced. Combined with shot change rate and sound energy, a novel united framework of excitement modeling and video abstract was proposed. First, the exciting content of a video was modeled and the excitement time curve was obtained. Then, according to local maximum and minimum,video abstracts of two forms （i. e. key frames and highlights） were extracted. Experiments confirm the ef- fectiveness of the modeling method. Video content can be represented effectively by the extracted abstract. This method shows 75% of key frames effectiveness, and, 81% and 82% of highlights effectiveness and ranking effectiveness respectively on the average over soccer video, and, 75% and 76% over basketball. Moreover, this approach shows good user-oriented and adaptive performance.