| 安文志,冯宇平,李云文,赵军,董金宇.基于多尺度图卷积网络的骨架行为识别方法[J].高技术通讯(中文),2025,35(1):37~46 |
| 基于多尺度图卷积网络的骨架行为识别方法 |
| Multi-scale graph convolution network for skeleton-based action recognition |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 01. 004 |
| 中文关键词: 骨架行为识别; 多尺度; 图卷积; 空间通道注意力 |
| 英文关键词: skeleton-based action recognition, multi-scale, graph convolution, spatial channel attention |
| 基金项目: |
| 作者 | 单位 | | 安文志 | (青岛科技大学自动化与电子工程学院 青岛 266061) | | 冯宇平 | | | 李云文 | | | 赵军 | | | 董金宇 | |
|
| 摘要点击次数: 197 |
| 全文下载次数: 226 |
| 中文摘要: |
| 针对现有基于骨架行为识别的方法缺乏对远距离节点之间关系的建模从而导致识别准确率低和泛化能力差的问题,提出一种基于多尺度图卷积网络(multi-scale graphconvolutional metwork,MS-GCN)的行为识别方法。 首先,通过多跳邻接矩阵实现扩张图卷积的构建,并结合不同跳数的扩张图卷积构建多尺度空间图卷积;其次,提出空间通道注意力(spatial channel attention,SCA)以激发空间敏感通道进一步增强空间特征;最后,采用均匀采样的数据增强方式生成多样化的训练样本,增强模型的鲁棒性和泛化能力。 所提方法在数据集 NTU-RGB + D 60、NTU-RGB + D 120 和 Northwestern-UCLA 上准确率分别达到了 97. 24% (X-View)、90. 43% (X-Set)和 96. 34% ,验证了该方法的有效性。 |
| 英文摘要: |
| In view of the lack of modeling of the relationship between remote nodes in existing skeleton-based action recognition methods, which leads to low recognition accuracy and poor generalization ability, a new action recognition method based on multi-scale graph convolutional network (MS-GCN) is proposed. Firstly, the convolution of extended graphs is constructed by multi-hop adjacency matrix, and the convolution of multi-scale spatial graphs is constructed by combining the convolution of expanded graphs with different hop numbers. Secondly, spatial channel attention (SCA) is proposed to stimulate spatial sensitive channels to further enhance spatial features. Finally, the uniform sampling data enhancement method is used to generate diversified training samples to enhance the robustness and generalization ability of the model. The accuracy of the proposed method is 97.24%(X-View), 90.43%(X-Set) and 96.34% on the data sets NTU-RGB+D 60, NTU-RGB+D 120 and Northwestern-UCLA, respectively, which fully verifies the effectiveness of the proposed method. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
