Related Experiment Video
Updated: Jun 12, 2025
An Inertial Measurement Unit Based Method to Estimate Hip and Knee Joint Kinematics in Team Sport Athletes on the Field
Published on: May 26, 2020
Real-Time Estimation of Lower Limb Posture and Joint Angles Using Wearable IMUs: Reproduced Hemiparetic and Normal
Shun Kanega1,2, Yoshihiro Muraoka1,2
1Faculty of Human Sciences, Waseda University, Tokorozawa, Saitama, Japan.
Abstract:
Kanega S, Muraoka Y. Real-Time Estimation of Lower Limb Posture and Joint Angles Using Wearable IMUs: Reproduced Hemiparetic and Normal Gait. Jpn J Compr Rehabil Sci 2025; 16: 19-29.
Objective:
A low-computational-load posture estimation method was employed, assuming integration into a functional electrical stimulation (FES) device, to estimate thigh and shank inclination angles as well as the knee joint angle during gait. The accuracy and processing time of the method were evaluated to assess its practical feasibility.
Methods:
Two inertial measurement units (IMUs) were mounted on the lateral thigh and shank. Two healthy adults performed level-ground gait, including normal gait and gait reproducing the characteristics of circumduction gait, which is typically observed in individuals with post-stroke hemiparesis. A custom gait pacemaker was developed to replicate the spatiotemporal asymmetry of hemiparetic gait. Posture estimation in the sagittal and frontal planes was performed on a microcontroller using a low-computational-load Madgwick filter, and the processing time was recorded. The estimation accuracy was assessed by comparing the results with optical motion capture data using root mean square error (RMSE) and cross-correlation analysis.
Results:
The average total processing time, including sampling, posture estimation, and stimulation control, was 6.8 ms. The RMSE of sagittal plane posture angles was less than 4° in all cases, suggesting estimation accuracy comparable to previous studies. For the frontal plane posture angles, the estimation accuracy was relatively high during the reproduced circumduction gait, indicating that the system effectively captured the circumduction motion. Conversely, the accuracy tended to be lower during normal gait.
Conclusion:
Compared with previous studies, the posture estimation method used in this study, which employed the Madgwick filter, was able to estimate posture without compromising accuracy. The entire process, including sampling, posture estimation, and stimulation control, was completed within 10 ms, reflecting the feasibility of real-time processing at 100 Hz.
