基于局部语义的人体动作识别方法  

Local semantic concept-based human action recognition

作　　者：林红[1] 张静[1] 

机构地区：[1]天津大学电子信息工程学院,天津300072

出　　处：《信息技术》2015年第12期114-116,120,共4页Information Technology

摘　　要：人体动作可以由人体不同局部区域的动作语义的组合来描述,由此提出了一种基于局部语义的人体动作识别方法。首先,该方法定义了一组局部动作语义用于描述人体局部区域运动的视觉表现,并对每一个局部语义进行建模。然后,通过这些局部动作语义的贡献值组合来进行构建动作表征。最后,将基于局部动作语义的动作表征输入支持向量机构建动作模型,进行动作分类。比较实验说明所提出方法能够较好地识别现实场景下的人体动作。Human action can be described by a set of local semantic concepts of different human body regions. This paper proposes a human action recognition method based on local semantic concept. First, it defines a set of local semantic concepts corresponding to the visual characteristics of local human body regions＇ motion and model these semantic concepts. Then, the action representation is built based on the contribution score of each semantic concept. Final, this local semantic concept based action representations is used for action classification with SVM classifier. The experiment demonstrates the proposed method can well recognize human action in the real scene.

关 键 词：动作识别 局部动作语义 支持向量机 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]