基于乘积格拉斯曼流形的人体骨架动作识别  

HUMAN SKELETON ACTION RECOGNITION BASED ON PRODUCT GRASSMANN MANIFOLD

作　　者：林枫 Lin Feng(School of Computer Science and Technology,University of Science and Technology of China,Hefei 230027,Anhui,China)

机构地区：[1]中国科学技术大学计算机科学与技术学院,安徽合肥230027

出　　处：《计算机应用与软件》2022年第8期213-219,共7页Computer Applications and Software

基　　金：自然资源部项目“自然资源标准语义一致性分析研究”。

摘　　要：3D人体骨架识别是近年来计算机视觉领域的研究热点。为了对具有非线性流形结构的高维股价序列数据进行准确而高效的分类,提出一种基于乘积格拉斯曼流形的人体骨架动作识别方法。利用乘积格拉斯曼流形描述序列数据的流形结构,构造出一种新的低维子空间表示,并借助流形上的非线性度量比较序列之间的差异。在此基础上,提出一种融合时序信息的核,从而实现对动作序列的高效分类。实验结果表明,该方法在多个骨架动作数据集上都能提高动作识别的准确率,验证了该方法的有效性,进而为更广泛的时间序列分类提供了新思路。In recent years,3 D human skeleton action recognition has become a hot research topic in computer vision.In order to accurately and efficiently classify the high-dimensional skeleton sequences with non-linear manifold structure,this paper proposes a method for human skeleton action recognition based on product Grassmann manifold.We used the product Grassmann manifold to model the manifold structure of the sequential data,which yielded a new low-dimensional subspace representation.The non-linear metrics on the manifold were employed to measure the difference between sequences.On this basis,a type of kernels that contained temporal order information was proposed to achieve efficient classification for action sequences.Experimental results show that the method can improve the accuracy of action recognition on several skeleton action datasets,which verifies its effectiveness.It further provides a new idea for more extensive time series classification.

关 键 词：动作识别 乘积格拉斯曼流形 子空间 时序信息 核 

分 类 号：TP[自动化与计算机技术]