An Auto-Calibrating Knee Flexion-Extension Axis Estimator Using Principal Component Analysis with Inertial Sensors.

Timothy Mcgrath,Leia Stirling,Richard Fineman

doi:10.3390/s18061882

Abstract

Inertial measurement units (IMUs) have been demonstrated to reliably measure human joint angles—an essential quantity in the study of biomechanics. However, most previous literature proposed IMU-based joint angle measurement systems that required manual alignment or prescribed calibration motions. This paper presents a simple, physically-intuitive method for IMU-based measurement of the knee flexion/extension angle in gait without requiring alignment or discrete calibration, based on computationally-efficient and easy-to-implement Principle Component Analysis (PCA). The method is compared against an optical motion capture knee flexion/extension angle modeled through OpenSim. The method is evaluated using both measured and simulated IMU data in an observational study (n = 15) with an absolute root-mean-square-error (RMSE) of 9.24 and a zero-mean RMSE of 3.49. Variation in error across subjects was found, made emergent by the larger subject population than previous literature considers. Finally, the paper presents an explanatory model of RMSE on IMU mounting location. The observational data suggest that RMSE of the method is a function of thigh IMU perturbation and axis estimation quality. However, the effect size for these parameters is small in comparison to potential gains from improved IMU orientation estimations. Results also highlight the need to set relevant datums from which to interpret joint angles for both truth references and estimated data.

Highlights

IntroductionAccurate measurement of human joint angles is central to the study of human biomechanics
Accurate measurement of human joint angles is central to the study of human biomechanics.Better biomechanical models and measurement systems enable more robust tools for interacting with and understanding human kinematics
Bland–Altman analysis of the total experiment in Figure 5a shows a linearity of r2 = 0.80 to the best-fit model y = 1.04x − 0.32, a 95% confidence interval of the error on [+20,−18]◦, and an average static bias offset of 0.63◦ (p < 0.001)

Summary

Introduction

Accurate measurement of human joint angles is central to the study of human biomechanics. Better biomechanical models and measurement systems enable more robust tools for interacting with and understanding human kinematics. For sports applications, understanding motion can lead to improved strategy development. To this end, an accurate measurement of human joint angles is desired. An accurate measurement of human joint angles is desired This measurement is complicated, as human motion is characteristically nonlinear, non-smooth, and uncorrelated in time [1]. Optical motion capture is accurate in triangulating reflective marker position in space, but interpretation of these data as human joint angles requires an assumed human model. From markers set on major anatomical landmarks of the body, a least-squares optimization to fit a model may be performed to estimate the joint angles of interest

Methods

Results

Conclusion