[1]张 艳,肖文琛,张 博.基于双流骨架信息的人体动作识别方法[J].计算机技术与发展,2024,34(01):158-163.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 023]
　ZHANG Yan,XIAO Wen-chen,ZHANG Bo.Human Action Recognition Method Based on Two-flow Skeleton Information[J].,2024,34(01):158-163.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 023]

点击复制

基于双流骨架信息的人体动作识别方法()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年01期

页码:

158-163

栏目:

人工智能

出版日期:

2024-01-10

文章信息/Info

Title:

Human Action Recognition Method Based on Two-flow Skeleton Information

文章编号:

1673-629X(2024)01-0158-06

作者:

张 艳; 肖文琛; 张 博

北华航天工业学院 计算机学院,河北 廊坊 065000

Author(s):

ZHANG Yan; XIAO Wen-chen; ZHANG Bo

School of Computing,North China Institute of Aerospace Engineering,Langfang 065000,China

关键词:

动作识别; 骨架信息; 注意力机制; 图卷积神经网络; 双流网络

Keywords:

action recognition; skeleton information; attention mechanism; graph convolutional neural network; two-stream networks

分类号:

TP391. 4

DOI:

10. 3969 / j. issn. 1673-629X. 2024. 01. 023

摘要:

针对当前基于二维图像的人体动作识别算法鲁棒性差、识别率不高等问题,提出了一种融合卷积神经网络和图卷积神经网络的双流人体动作识别算法,从人体骨架信息提取动作的时间与空间特征进行人体动作识别。 首先,构建人体骨架信息时空图,利用引入注意机制的图卷积网络提取骨架信息的时间和空间特征;其次,构建骨架信息运动图,将卷积神经网络网络提取到骨架运动信息的特征作为时空图卷积网络所提取特征的时间和空间特征的补充;最后,将双流网络进行融合,形成基于双流的、注意力机制的人体动作识别算法。 算法增强了骨架信息的表征能力,有效提高了人体动作的识别精度,在 NTU -RGB+D60 数据集上取得了比较好的结果,Cross-Subject 和 Cross-View 的识别率分别为 86. 5% 和93. 5% ,相比其他同类算法有一定的提高。

Abstract:

Aiming at the problems of poor robustness and low recognition rate of current human action recognition algorithms based ontwo- dimensional images, a two - stream human?

action recognition algorithm based on convolutional neural network and graphconvolutional neural network was proposed to extract the temporal and spatial features of human?

action recognition from human skeletoninformation. Firstly, the spatial and temporal graph of skeleton information is constructed, and the graph convolution network withattention mechanism is used to extract the temporal and spatial characteristics of skeleton information. Secondly,the skeleton informationaction graph is constructed,and the features extracted from the convolutional neural network are used as the time and space features of thefeatures extracted from the spatio-temporal graph convolutional network. Finally,

the two-stream networks are fused to form a humanaction recognition algorithm based on dual flow and attention mechanism. The proposed algorithm enhances the representation ability ofskeleton information and effectively improves the recognition accuracy of human movements. It achieves good results on the NTU-RGB+D60 data set,and the recognition rates of Cross - Subject and Cross - View are 86. 5% and 93. 5% , respectively, which is a certainimprovement compared with other similar algorithms.

常用功能