Physical Non-inertial Poser (PNP): Modeling Non-inertial Effects in Sparse-inertial Human Motion Capture

• DOI: 10.48550/arxiv.2404.19619
• Publication Date: 2024-04-30
• Authors: Xinyu Yi; Yuxiao Zhou; Feng Xu
• Keywords: Inertial reference unit

Existing inertial motion capture techniques use the human root coordinate frame to estimate local poses and treat it as an by default. We argue that when has linear acceleration or rotation, should be considered non-inertial theoretically. In this paper, we model fictitious forces are non-neglectable in a auto-regressive estimator delicately designed following physics. With forces, force-related IMU measurement (accelerations) can correctly compensated thus Newton's laws of satisfied. case, relationship between accelerations body motions is deterministic learnable, train neural network for better capture. Furthermore, with synthetic data, develop synthesis simulation strategy noise hardware allow parameter tuning fit different hardware. This not only establishes training data but also enables calibration error modeling handle bad calibration, increasing robustness system. Code available at https://xinyu-yi.github.io/PNP/.