@

초록

Inertial measurement units (IMUs) have been used for motion analysis, but accurate joint angle estimation requires alignment between each sensor frame and the anatomical segment. To avoid explicit sensor– segment alignment, we estimate the joint axis directly from data. We introduce PCA-3S, principal component analysis (PCA)–based joint axis estimation with sample selection scheme that extracts single-degree-of-freedom (DoF) segments from multi-DoF-contaminated calibration data. PCA-3S is designed to address the limitations of conventional PCA methods, which are sensitive to multi-DoF motion contamination during calibration. Using the estimated axes, joint angles are obtained via an axis-based rotation matrix decomposition that reduces calibration burden.We evaluate the method on an elbow apparatus and human subject data under varying contamination levels, speeds, and sensor misalignments. PCA-3S achieves consistent axis errors of 6.1° for flexion/extension and 3.1° for supination/pronation on the apparatus, with joint angle root mean square errors (RMSEs) below 5.0°. Human elbow experiments yielded RMSEs of approximately 8.5°, outperforming conventional PCA and optimization approaches. These results indicate that PCA-3S provides robust, accurate joint axis and joint angle estimation suitable for unconstrained environments.

키워드