Accuracy and Reliability of Monocular 3D Pose Estimation for Lower-Limb Movement Analysis
International Journal of Advanced Computer Science and Applications · 2026 · 17 (10)
DOI: 10.14569/ijacsa.2026.0171065 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Advanced Computer Science and Applications, 2026. Verfasst von Guangjun Wang und 5 weiteren Autor:innen.
Abstract (Originalsprache)
Optical motion capture provides accurate joint kinematics, but its cost and setup complexity limit routine use in rehabilitation and home monitoring. Monocular RGB video is a low-cost alternative, yet many 3D pose estimation studies report overall errors on public benchmarks without distinguishing lower-limb movements or individual joints. This study evaluated the spatial and frame-to-frame displacement errors of a two-stage pipeline combining BlazePose 2D detection with a dilated temporal convolutional network for 3D lifting. We constructed a paired dataset of monocular RGB video and optical motion capture from 10 healthy participants performing six standardized lower-limb movements, yielding 450 samples and approximately 202,500 frames. Under the within-participant recording-set evaluation, the fine-tuned pipeline achieved an MPJPE of 32.03 mm, compared with 121.70 mm for direct BlazePose 3D output. Errors varied across movements, from 11.21 mm for quiet standing to 62.83 mm for seated knee extension, with distinct patterns at the hip, knee, and ankle. At a noise standard deviation of 5.4 pixels, independent Gaussian perturbations of 2D keypoints increased MPJPE by approximately 0.9% but increased mean per-joint velocity error (MPJVE), measured here as frame-to-frame displacement error, by about 234%. Selected joint-angle measurements also showed movement-dependent agreement with optical motion capture. These findings provide movement- and joint-specific evidence on position accuracy, temporal sensitivity, and selected kinematic agreement under controlled laboratory conditions.
Zitieren
APA
Wang, G., Chen, J., Hu, M., Luo, C., Li, K., & Chen, X. (2026). Accuracy and Reliability of Monocular 3D Pose Estimation for Lower-Limb Movement Analysis. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/ijacsa.2026.0171065
Vancouver
Wang G, Chen J, Hu M, Luo C, Li K, Chen X. Accuracy and Reliability of Monocular 3D Pose Estimation for Lower-Limb Movement Analysis. International Journal of Advanced Computer Science and Applications. 2026;17(10). doi:10.14569/ijacsa.2026.0171065
BibTeX
@article{Wang2026,
author = {Wang, Guangjun and Chen, Jiaojiao and Hu, Mengxia and Luo, Cezhong and Li, Kaibo and Chen, Xiaohui},
title = {Accuracy and Reliability of Monocular 3D Pose Estimation for Lower-Limb Movement Analysis},
journal = {International Journal of Advanced Computer Science and Applications},
year = {2026},
volume = {17},
number = {10},
doi = {10.14569/ijacsa.2026.0171065}
}
EUVIMED