基于高层语义词袋的人体行为识别方法

Human Activity Recognition Based on High-level Semantic Codebook

人体行为分析为视频监控系统、视频检索系统提供重要的研究基础。本文提出了一种基于高层语义词袋模型的人体行为识别方法。该方法根据底层词袋中词汇的相关关系,构造出一个基于词汇交互信息量的底层词汇图;然后使用层次聚类的方法对该图进行分割,得到底层词汇组模型,最后将该模型表示为高层语义词袋模型。实验结果表明,该方法可以高效地识别视频中的人体行为。

Human activity recognition is the important basis of video surveillance system. In this paper, a new activity recognition method is proposed based on the high-level codebook. We construct a code-word graph based on the mutual information of lowlevel code-words, and then partition the graph into different groups, which discover the high-level code-words patterns. Experimental result shows that the proposed method can effective recognize human activities.