中文摘要：

文章主要研究基于视频的行为识别技术．．该技术主要用于识别时空混合的视频数据中的人类行为活动，它不同于二维图像识别技术和三维时空兴趣点检测方法。主要讨论如何通过提取立方体像素窗口、聚类生成特征字典完成识别工作。文章所使用的方法主要包括离散线性过滤器的实现和关键参数的选择、对K—means聚类算法的改进、分类器设计等.通过在公开数据库上的实验，表明文章使用的方法是有效的，取得了比较理想的结果、

英文摘要：

This paper focuses on the behavior recognition technology video. The technology is mainly used for activities of human behavior recognition time video data in hybrid, which is different from the two-dimensional image recognition technology and 3D spatio-temporal interest point detection method. Mainly discuss how to extract the cube pixel window, clustering feature dictionary complete identification. The method used by the authors mainly includes the implementation and key parameters of discrete linear filter selection, improvement of K-means clustering algorithm, the design of classifier. Through the open database experiments, show that this method is effective, and the ideal results are achieved.