计算机应用研究2025，Vol.42Issue(11)：3497-3505,9.DOI:10.19734/j.issn.1001-3695.2025.03.0056
多分支骨架特征输入下的歧义行为识别
Ambiguous action recognition under multi-input-branch skeleton feature
摘要
Abstract
Skeleton-based human action recognition is a critical research topic in computer vision,focusing on extracting and learning discriminative skeletal features to achieve high-precision action classification.However,the presence of ambiguous ac-tions severely impacts classification accuracy.To address these challenges,this paper introduced two key innovations based on data optimization and computational complexity reduction as well as spatio-temporal feature refinement.Firstly,it employed a multiple-input-branches architecture in the early stage of data processing to facilitate early feature fusion,enabling the model to learn complementary information across different modalities more effectively.This design enhanced computational efficiency while improving the model's generalization ability.Secondly,to improve the recognition of highly similar actions,it proposed an ambig-uous-feature-refinement module to extract distinctive spatio-temporal features.This mechanism enhanced the model's sensitivity to action details,thereby achieving more refined spatio-temporal feature modeling.Finally,it evaluated the proposed ambiguous action recognition under multi-input-branch skeleton feature model(GCN+)on two large-scale public datasets,D60 and D120,covering four single-modal settings as well as their fused modalities.Experimental results demonstrate that:in single-modal set-tings,GCN+outperforms the baseline models,achieving a 2.6 percentage points accuracy improvement under the X-Sub eva-luation protocol on the D120 dataset,indicating superior robustness in recognizing actions across different subjects in complex en-vironments.In fused-modal settings,GCN+achieves a 3.2 percentage points increase in X-Sub accuracy and a 3.0 percentage points increase in X-Set accuracy on the D120 dataset,further validating its applicability in large-scale data scenarios and its out-standing performance in cross-subject and cross-view action recognition tasks.Overall,the experimental results confirm that GCN+exhibits strong generalization capability,high computational efficiency,and exceptional performance in ambiguous highly simi-lar actions,providing an effective and robust solution for skeleton-based action recognition in complex environments.关键词
骨架识别/图卷积神经网络/多分支输入/歧义行为/模糊特征细化Key words
skeleton recognition/graph convolution network/multiple-input-branches/ambiguous action/ambiguous fea-ture refinement分类
计算机与自动化引用本文复制引用
王超亚,韩华,王春媛,田瑾..多分支骨架特征输入下的歧义行为识别[J].计算机应用研究,2025,42(11):3497-3505,9.基金项目
国家自然科学基金项目(62103257) （62103257）
上海市自然科学基金项目(22ZR1426200) （22ZR1426200）
