Accuracy of Imus and Monocular Human Pose Estimation at Measuring Elbow Flexion

Markerless human pose estimation and inertial measurement units are two common methods for human movement analysis. They are cost-effective and easy to use, but their accuracy compared to established optical motion capture methods is yet to be determined for some movements. This study compared the accuracy of IMUs and BlazePose, a markerless monocular 3D human pose estimator, against a marker-based optical system for measuring elbow flexion angles. After performing a static offset calibration to account for differences in anatomical coordinate systems, these were reduced to a mean absolute error of 4.2° and a root mean square error of 6.0°. The accuracy of markerless pose estimation was significantly influenced by the orientation of the participants to the camera. The optimal camera orientation was achieved when participants performed the movement fully within the image plane, with a mean absolute error of 10.5° and a root mean square error of 14.7° after performing a static offset calibration. Bland-Altman analysis indicated mostly constant deviations for the IMUs but non-constant deviations for HPE, which underestimated large elbow flexion angles. The findings of this study underscore the importance of camera orientation for monocular HPE accuracy. IMUs display much higher accuracy at measuring elbow flexion than monocular HPE. Further improvements in the accuracy of HPE independent of orientation might broaden its applications in motion analysis.