一种融合隐藏语义的人体行为表示方法

An Action Representation Method by Fuse Hidden Semantic

人体行为识别是计算机视觉研究的热门领域之一,提出了一种基于隐藏语义的人体行为算法,采取人体骨骼点的三维数据进行处理后构成数据序列,将该数据序列作为分析人体行为的主语义的同时,分析人体行为细节信息在时空特性上的变化作为隐藏语义,然后将二者融合作为人体行为表示.最后使用改进的类均值核主成分分析算法对行为表示数据进行处理,并用支持向量机进行分类.将提出的方法在UTKinect、Florence和MSR Action 3D数据集上进行验证,实验结果证明了所提方法的有效性和普适性.

Human action recognition is one of the most popular research fields of computer vision research.This paper presents a human action representation algorithm based on the hidden semantics,this algorithm processes the three-dimensional data of the skeleton points to constitutes a data sequence,and the data sequence is used as the main semantics for analyzing human behavior,analyzes the changes of human action details in the temporal and spatial characteristics as the hidden semantics,after that,mixes the two data as human action representation.Finally use the improved CMKPCA algorithm to deal with the action representation data,use SVM for classification.This method is validated on UTKinect,Florence and MSR Action 3D datasets.The experimental results show the effectiveness and universality of the proposed method.