基于时空与或图模型的视频人体动作识别方法  

Video Human Action Recognition Based on Spatial Temporal And-Or Graph Model

作　　者：易唐唐[1] 

机构地区：[1]湖南女子学院信息技术系,长沙410004

出　　处：《控制工程》2017年第9期1792-1797,共6页Control Engineering of China

基　　金：湖南省教育厅科学研究青年基金资助项目(12B066)

摘　　要：针对视频中的人体动作识别问题,提出一种基于时空与或图(AOG)模型的人体动作识别方法,将动作识别和姿态估计共同建模在一个时空AOG模型中。首先,将动作分解为姿态,进一步将姿态分解为多个时空(ST)部件,再将ST-部件分解为多个子部件,从而形成三层模型;然后,在三层上分别提取粗级、中级和细级特征;最后,分别学习3个级别上的模型参数,训练隐含参数支持向量机(Latent SVM)分类器,实现动作识别。通过大型数据集的测试以及与几种最新方法的比较,证明了该方案的有效性,识别精度能够达到94%左右。For the issues of the human action recognition in video, a human motion recognition scheme based on spatial temporal AOG model is proposed, in which the action recognition and pose estimation are modeled in a AOG model. First, the action is decomposed into some poses, then the poses are decomposed into multiple spatial temporal （ST） parts, and the ST-parts are decomposed into multiple sub parts, so as to form the three-layer model. Then, the coarse-level, mid-level and fine-level features are extracted respectively. Finally, the model parameters from three levels are learned to train the latent parameters support vector machine （Latent SVM） classifier, and finally realizes the action recognition. Experiments are carried out on several data sets, and compared with several new methods. The experimental results demonstrate the effectiveness of the scheme, and the recognition accuracy can reach about 94 %.

关 键 词：人体动作识别 与或图模型 姿态估计 隐含参数支持向量机 时空特征学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]