| 摘 要: 地面反作用力(Ground Reaction Force，GRF)、地面反作用力矩(Ground Reaction Moment，GRM)和压力中心(Center of Pressure，COP)是人体运动生物力学分析的核心参数，在临床步态评估、运动损伤预防和外骨骼控制中具有重要应用价值。传统测力台受场地与成本限制，难以满足真实场景连续监测需求。系统综述了2018至2025年间基于可穿戴传感器与深度学习预测GRF/GRM/COP的研究进展，涵盖传感器类型与数据处理流程，并从递归神经网络、卷积与时序卷积网络、注意力机制与Transformer及混合架构四个维度对代表性方法进行对比分析。研究表明，混合架构在多维联合预测中表现突出，少量个体数据的自适应校准可显著降低跨个体误差。最后，针对多运动类型泛化、跨个体迁移和轻量化部署等挑战，展望了物理先验融合与基础模型预训练等研究方向。 |
| 关键词: 地面反作用力 可穿戴传感器 深度学习 步态分析 混合架构 |
|
中图分类号:
文献标识码:
|
| 基金项目: 浙江省重点研发计划项目(2024C03055)；浙江省自然科学基金项目(25222261D)；大学生创新创业训练计划项目(S202510338082) |
|
| A Survey of Deep Learning Methods for Predicting Human Movement Biomechanical Parameters from Wearable Sensors |
|
yu ming, wang chen, fan gaoyang, chen hao, dai yanyun
|
School of Information Science and Engineering (School of Cyberspace Security), Zhejiang Sci-Tech University
|
| Abstract: Ground reaction force(GRF), ground reaction moment(GRM), and center of pressure(COP) are core parameters in human movement biomechanics, with important applications in clinical gait assessment, sports injury prevention, and exoskeleton control. However, conventional force plates are constrained by fixed measurement areas and high costs, limiting continuous real-world monitoring. This paper systematically reviews research from 2018 to 2025 on predicting GRF/GRM/COP using wearable sensors combined with deep learning, covering sensor types and data processing pipelines, and comparing representative methods across four dimensions: recurrent neural networks, convolutional and temporal convolutional networks, attention mechanisms with Transformers, and hybrid architectures. Results indicate that hybrid architectures excel in multi-dimensional joint prediction, and subject-adaptive calibration with limited individual data can significantly reduce cross-subject errors. Finally, future research directions including physics-informed learning and foundation model pretraining are discussed to address key challenges of multi-activity generalization, cross-subject transfer, and lightweight deployment. |
| Keywords: ground reaction force wearable sensors deep learning gait analysis hybrid architecture |
