| 31 | 0 | 94 |
| 下载次数 | 被引频次 | 阅读次数 |
现有的基于通道拓扑细化图卷积网络的人体骨架动作识别方法在关节特征与通道拓扑建模中缺乏有效筛选机制,导致关键局部特征不突出及重要拓扑信息易被弱化。针对此问题,提出一种轻量化增强型人体骨架动作识别方法。首先,对原始时空图卷积单元进行改进,在其内部引入高效通道注意力机制,以提升特征表达能力;其次,设计基于Top-K显著响应引导的特征聚合策略,在保留全局拓扑信息的同时强化局部显著特征表示,从而增强模型对关键动作模式的刻画能力。在NTU RGB+D 60和NTU RGB+D 120数据集上的实验结果表明,所提方法在几乎不增加参数量与计算复杂度的前提下有效提升动作识别准确率,在精度与效率之间取得了平衡。
Abstract：Existing skeleton action recognition methods based on channel-wise topology refinement graph convolutional networks lacked effective screening mechanisms in joint feature and channel-wise topology modeling, resulting in insufficient prominence of key local features and easy attenuation of important topological information. To address this issue, a lightweight enhanced human skeleton action recognition method was proposed. Firstly, the original spatiotemporal graph convolution unit was improved by introducing an efficient channel attention mechanism to enhance its feature expression ability. Secondly, a Top-K salient response-guided feature aggregation strategy was designed to enhance local salient feature representations while preserving global topological information, thereby improving the model′s ability to capture discriminative action patterns. The experimental results on the NTU RGB+D 60 and NTU RGB+D 120 datasets showed that the proposed method could effectively improve the accuracy of action recognition, with almost no increase in the number of parameters and computational complexity, achieving a balance between accuracy and efficiency.
[1]Song S J, Lan C L, Xing J L, et al. An end-to-end spatio-temporal attention model for human action recognition from skeleton data[C]∥Proceedings of the 31st AAAI Conference on Artificial Intelligence. Palo Alto:AAAI Press,2017:4263-4270.
[2]Gao Y B, Li C K, Li S, et al. Variable rate independently recurrent neural network(IndRNN)for action recognition[J]. Applied Sciences, 2022, 12(7):3281.
[3]Ke Q H, Bennamoun M, An S J, et al. A new representation of skeleton sequences for 3D action recognition[C]∥IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2017:4570-4579.
[4]Liu M Y, Liu H, Chen C. Enhanced skeleton visualization for view invariant human action recognition[J]. Pattern Recognition, 2017, 68:346-362.
[5]Yan S J, Xiong Y J, Lin D H. Spatial temporal graph convolutional networks for skeleton-based action recognition[C]∥Proceedings of the 32nd AAAI Conference on Artificial Intelligence. Palo Alto:AAAI Press, 2018:7444-7452.
[6]高晗,万方杰,马明旭.结合ResNet和CBAM的静态图像行为识别方法[J].郑州大学学报(理学版),2025, 57(3):65-71.Gao H, Wan F J, Ma M X. Still image action recognition method combining ResNet and CBAM[J]. Journal of Zhengzhou University(Natural Science Edition), 2025,57(3):65-71.
[7]Xiang W M, Li C, Zhou Y X, et al. Generative action description prompts for skeleton-based action recognition[C]∥IEEE/CVF International Conference on Computer Vision. Piscataway:IEEE Press, 2023:10242-10251.
[8]Chi H G, Ha M H, Chi S, et al. InfoGCN:representation learning for human skeleton-based action recognition[C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2022:20154-20164.
[9]Lee J, Lee M, Lee D, et al. Hierarchically decomposed graph convolutional networks for skeleton-based action recognition[C]∥IEEE/CVF International Conference on Computer Vision. Piscataway:IEEE Press, 2023:10410-10419.
[10]Lee J, Lee M, Cho S, et al. Leveraging spatio-temporal dependency for skeleton-based action recognition[C]∥IEEE/CVF International Conference on Computer Vision.Piscataway:IEEE Press, 2023:10221-10230.
[11]Zhou Y X, Yan X D, Cheng Z Q, et al. BlockGCN:redefine topology awareness for skeleton-based action recognition[C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2024:2049-2058.
[12]Liu H D, Liu Y F, Ren M, et al. Revealing key details to see differences:a novel prototypical perspective for skeleton-based action recognition[C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2025:29248-29257.
[13]ChenYX,ZhangZQ,YuanCF,etal.Channel-wise topologyrefinementgraphconvolutionforskeleton-based/actionrecognition[C]∥IEEE CVFInternationalConferenceonComputerVision.Piscataway:IEEEPress,2021:13339-13348.
[14]LiuZY,ZhangHW,ChenZH,etal.Disentangling andunifyinggraphconvolutionsforskeleton-basedaction/recognition[C]∥IEEE CVFConferenceonComputerVisionandPatternRecognition.Piscataway:IEEEPress,2020:140-149.
[15]Shahroudy A, Liu J, Ng T T, et al. NTU RGB+D:a large scale dataset for 3D human activity analysis[C]∥IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press,2016:1010-1019.
[16]Liu J, Shahroudy A, Perez M, et al. NTU RGB+D 120:a large-scale benchmark for 3D human activity understanding[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(10):2684-2701.
[17]Cheng K, Zhang Y F, Cao C Q, et al. Decoupling GCN with DropGraph module for skeleton-based action recognition[C]∥Proceedings of the European Conference on Computer Vision. Cham:Springer International Publishing, 2020:536-553.
[18]Zhou H Y, Liu Q J, Wang Y H. Learning discriminative representations for skeleton based action recognition[C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2023:10608-10617.
基本信息:
DOI：10.13705/j.issn.1671-6841.2026020
中图分类号:TP391.41;TP183
引用信息:
[1]罗勇,张鹏远.基于高效通道注意力的骨架动作识别方法[J].郑州大学学报(理学版)().DOI:10.13705/j.issn.1671-6841.2026020.
基金信息:
国家自然科学基金面上项目(62476254)
2026-09-09
2026-09-09
2026-09-09