Scholarlink

<p>The estimation of lower-extremity multi-joint kinematics is essential for biomechanical analysis and clinical purposes. The current standard for measuring the joint angle is using an optical motion capture system or several wearable inertial measurement unit sensors. However, the motion capture system is limited to laboratory environments, and using too many inertial measurement unit sensors can cause discomfort to the wearer. Hence, conducting research on accurately estimating lower-extremity multi-joint kinematics using a limited number of inertial measurement units becomes imperative. In this work, we propose a novel deep learning model that combines modified Informer encoder and temporal convolutional neural network to estimate lower-extremity joint angles (hip, knee, and ankle) in the sagittal plane. Gated recurrent unit, long short-term memory, and other mainstream methods are adopted as baselines for comparison on three publicly datasets. Root mean square error, mean absolute error, coefficient of determination, and Pearson correlation coefficient are chosen as evaluation metrics. The results show that the proposed model outperforms other methods when estimating lower-extremity multi-joint kinematics with only a single inertial measurement unit in most cases. Meanwhile, combining modified Informer encoder with other deep learning methods can also improve the estimation performance. The proposed method can effectively extract the temporal characteristics of the signals and capture the long-term dependencies of feature sequences. Moreover, this study offer a practical, low-burden solution for clinical gait analysis and rehabilitation monitoring, reducing sensor complexity and calibration overhead while maintaining accurate multi-joint kinematics estimation. The code is released on GitHub at <a href="https://github.com/Shurun-Wang/MIE-TCN">https://github.com/Shurun-Wang/MIE-TCN</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Estimating Lower-Extremity Multi-Joint Kinematics with One IMU Sensor via Attention-based Temporal Convolutional Neural Network

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • Shurun Wang,
  • Hao Tang,
  • Ryutaro Himeno,
  • Jordi Solé-Casals,
  • Ying Tan,
  • Jie Mao,
  • Cesar Caiafa,
  • Zhe Sun

摘要

The estimation of lower-extremity multi-joint kinematics is essential for biomechanical analysis and clinical purposes. The current standard for measuring the joint angle is using an optical motion capture system or several wearable inertial measurement unit sensors. However, the motion capture system is limited to laboratory environments, and using too many inertial measurement unit sensors can cause discomfort to the wearer. Hence, conducting research on accurately estimating lower-extremity multi-joint kinematics using a limited number of inertial measurement units becomes imperative. In this work, we propose a novel deep learning model that combines modified Informer encoder and temporal convolutional neural network to estimate lower-extremity joint angles (hip, knee, and ankle) in the sagittal plane. Gated recurrent unit, long short-term memory, and other mainstream methods are adopted as baselines for comparison on three publicly datasets. Root mean square error, mean absolute error, coefficient of determination, and Pearson correlation coefficient are chosen as evaluation metrics. The results show that the proposed model outperforms other methods when estimating lower-extremity multi-joint kinematics with only a single inertial measurement unit in most cases. Meanwhile, combining modified Informer encoder with other deep learning methods can also improve the estimation performance. The proposed method can effectively extract the temporal characteristics of the signals and capture the long-term dependencies of feature sequences. Moreover, this study offer a practical, low-burden solution for clinical gait analysis and rehabilitation monitoring, reducing sensor complexity and calibration overhead while maintaining accurate multi-joint kinematics estimation. The code is released on GitHub at https://github.com/Shurun-Wang/MIE-TCN.