Scholarlink

This paper presents a low-cost and robust human pose estimation framework that fuses Inertial Measurement Units (IMUs) with Visual-inertial Odometry (VIO). A custom-designed wireless IMU module and distributed hardware architecture enable real-time estimation of 3D orientations using a quaternion-based Extended Kalman Filter (EKF). A hybrid static-dynamic alignment method is introduced to precisely map IMU frames to anatomical body frames. Based on this alignment, joint angles between adjacent body segments are computed, and the global body position is tracked via a pelvis-mounted VIO sensor. The proposed system is validated through visualization in the MuJoCo simulation platform, where full-body motion is accurately reproduced. Experimental results demonstrate the method’s high real-time performance and accuracy in reconstructing natural human motion, highlighting its applicability to wearable sensing and human-robot interaction.

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

A Low-Cost Multisensor IMU-VIO Framework for Real-Time Full-Body Human Pose Estimation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lele Li,
  • Zedong Liu,
  • Dawei Liang,
  • Chuanyu Si,
  • Haotian Ju,
  • Shouyi Zhang,
  • Haoxiang Zhang,
  • Hongwei Jing,
  • Jian Qi,
  • Tianjiao Zheng,
  • Yanhe Zhu

摘要

This paper presents a low-cost and robust human pose estimation framework that fuses Inertial Measurement Units (IMUs) with Visual-inertial Odometry (VIO). A custom-designed wireless IMU module and distributed hardware architecture enable real-time estimation of 3D orientations using a quaternion-based Extended Kalman Filter (EKF). A hybrid static-dynamic alignment method is introduced to precisely map IMU frames to anatomical body frames. Based on this alignment, joint angles between adjacent body segments are computed, and the global body position is tracked via a pelvis-mounted VIO sensor. The proposed system is validated through visualization in the MuJoCo simulation platform, where full-body motion is accurately reproduced. Experimental results demonstrate the method’s high real-time performance and accuracy in reconstructing natural human motion, highlighting its applicability to wearable sensing and human-robot interaction.