A Quantitative Comparison of Human Body Modeling Methods to Synthesize IMU Data
Year: 2026
Type: article
Abstract: In this work, we evaluate how accurately human body modeling approaches can reproduce inertial measurements from body-worn sensors. We focus on three body modeling approaches: 1) A personalized biomechanics-based skeletal model (BSK), 2) a personalized model that combines a biomechanical kinematic model with human surface model (BSU), and 3) a computer graphics model (SMPLX). All body models were independently created from anthropometric data, reference images, and motion capture data. For model evaluation, we simulated eight activities of daily living (ADL) that involve upper and lower limbs. Acceleration and angular velocity data were sampled from 14 different sensor positions on the body and compared against physical IMU measurements. Results show an average mean absolute error (MAE) of $13.0^{\circ } /s$ vs. $12.3^{\circ } /s$ vs. $13.5^{\circ } /s$ for angular velocity and $1.7{\,}m/s^{2}$ vs. $1.2{\,}m/s^{2}$ vs. $1.3{\,}m/s^{2}$ for acceleration of BSK, BSU and SMPLX, respectively. Mean correlation coefficients are between 0.73 to 0.83 for acceleration and between 0.78 to 0.82 for angular velocity when compared to physical IMU data. Spectral similarity, measured as mean Jensen-Shannon Distance (JSD), is ~0.11 for acceleration and angular velocity for all models Across sensor modalities, ADLs, and study participants (n = 19), the BSU model consistently showed lower mean errors, with an average MAE improvement of 18.4% compared to BSK and 10.0% compared to SMPLX. Overall, we conclude that the BSU model shows the most realistic representation of synthesized sensor signals, which is essential for simulation-based motion analysis, optimisation of wearable inertial sensor systems, data augmentation in human activity recognition tasks, joint kinematics prediction, and could enhance machine learning model robustness in wearable technology applications.
Source: IEEE Access
Language: English
Topic: Human Motion and Animation
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W7204118180