计算机应用研究2026，Vol.43Issue(3)：954-960,7.DOI:10.19734/j.issn.1001-3695.2025.05.0223
FDR-Net:融合动态图卷积与频域推理的人体运动预测
FDR-Net:human motion prediction combining dynamic graph convolution and frequency-domain reasoning
摘要
Abstract
With the rapid development of embodied intelligence and human-computer interaction technologies,accurate pre-diction of human motion intent becomes crucial for enhancing interaction naturalness and safety.To address the limitations of existing graph convolution-based human motion prediction methods in capturing spatio-temporal features and modeling long-term dependencies,this study proposed a human motion prediction method integrating dynamic graph convolution with frequency domain reasoning.The algorithm firstly designed a multi-hop maximal relative graph convolution to enhance the capability of capturing local motion patterns,combined with reparameterized conditional positional encoding to achieve position awareness.Secondly,it introduced a temporal-aware attention mechanism to model inter-frame dependencies.Finally,the method construc-ted a frequency-domain graph network to perform multi-scale motion feature reasoning in the frequency domain through Fourier transform.Comparative experiments on the Human3.6M and 3DPW datasets demonstrate that the proposed method achieves significantly lower average displacement error compared to state-of-the-art approaches,generating predictions closer to ground truth.These results verify that the integration of dynamic graph convolution and frequency-domain reasoning effectively en-hances the accuracy and long-term stability of human motion prediction,providing technical support for natural and safe human-computer interaction.关键词
人体运动预测/动态图卷积网络/频域图推理/重参数化/时序注意力/傅里叶变换Key words
human motion prediction/dynamic graph convolution/frequency graph reasoning/reparameterization/tempo-ral self-attention/Fourier transform分类
信息技术与安全科学引用本文复制引用
贾文博,张孙杰..FDR-Net:融合动态图卷积与频域推理的人体运动预测[J].计算机应用研究,2026,43(3):954-960,7.基金项目
国家自然科学基金资助项目(61603255) （61603255）
上海市晨光计划资助项目(18CG52) （18CG52）
