中文摘要：

为了从海量视频数据中快速提取感兴趣的信息，在研究、分析现有视频浓缩算法性能的基础上，提出了一种基于关键目标选择的视频浓缩方法。该方法用选择的代表性观测点组成的新的运动序列来代表目标原运动序列，从而消除了现有算法不能消除的内容上的冗余，进而提高视频浓缩效率；采用数据驱动方式进行关键观测点选择，通过把这一选择问题转换为最小描述长度（MDL）选取问题来实现自适应选择，从而克服现有算法在视频浓缩中因观测目标极多导致压缩效率下降和影响视觉效果的问题。通过在三个数据库上的试验，证明了该方法的有效性。

英文摘要：

To quickly find the useful information from a vast amount of video data, a novel video synops~s method based on key observation selection was presented after the analysis of the performance of existing video synopsis algo- rithms. The method uses the new motion sequence composed by the selected representative observations to represent targets＇ original motion sequence to eliminate the content redundancy existing video synosis algorithms can not eliminate, so the video synopsis efficiency can be improved. In addition, the method adopts a data-driven mode to select key observations, and achieves the adaptive selection by transforming the key observation selection into the optimization of the minimum description length （MDL） to overcome the synopsis efficiency degrading and the syn- opsis video confusing of esisting video synopsis algorithms caused by too many observations. The experiments on three real surveillance videos were conducted to validate the effectiveness of the proposed approach.