Evaluating Lifting Techniques: A Comparative Study of Kinematics and Dynamics Using IMU Sensors and the OpenCap Framework

This master’s thesis project was carried out during the stay at the Human Movement Bioengineering research group at Katholieke Universiteit in Leuven (Belgium), within the framework of the Erasmus+ program, thanks to the collaboration between the University of Padua and KU Leuven. Human motion analysis plays a crucial role in ergonomic assessment, particularly through tools such as MATE (Musculoskeletal-Modeling-Based, Full-Body Load-Assessment Tool for Ergonomists), which rely on accurate estimation of kinematics and dynamics. This thesis investigates the performance of OpenCap, a markerless video-based motion capture system, during three lifting techniques: squat, stoop, and torsion. The aim is to evaluate whether OpenCap can provide results comparable to those of inertial measurement units (IMUs) and thus represent a feasible alternative within the MATE framework. For this goal, kinematic and dynamic outcomes obtained with OpenCap were directly compared against those derived from Xsens IMUs. A comparative approach is adopted: motion is recorded simultaneously with both systems and processed in OpenSim to obtain joint kinematics and dynamics. Segmentation and temporal alignment procedures are applied, and the outputs are compared using RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), and Pearson’s correlation coefficient. Kinematic analysis includes both upper- and lower-body joints, while dynamics focuses on the upper body due to the absence of ground reaction force data. The findings show that lower-limb kinematics are generally captured more accurately than upper-limb ones, with knee and hip flexion achieving the highest agreement between systems. Upper-limb joints, especially elbows (> 25° in some cases) and arm rotations, present larger errors. Nevertheless, flexion–extension movements consistently show strong correlations (>0.8), even when RMSE values are high, indicating that the temporal trends are well reproduced. For dynamics, lumbar moments exhibit the largest errors, which are amplified by trunk inertia, but still display relatively high correlation values (up to 0.9). Across tasks, torsion lifting provides the best overall correlations, particularly in the transverse plane, highlighting the importance of range of motion in system performance. The study demonstrates that OpenCap can reproduce kinematic and dynamic patterns with accuracy comparable to IMUs for some degrees of freedom, especially in the sagittal plane. This suggests that OpenCap could have the potential to be integrated into the MATE tool as a less cumbersome, markerless acquisition system for ergonomic risk assessment. However, limitations remain for upper-limb joints and non-sagittal joint angles. Further research should include a larger sample size, the acquisition of ground reaction forces to extend the dynamic analysis to the lower limbs for an eventual complete analysis, and the integration of a marker-based system as a gold standard reference. Optimizing camera setup and testing different loads could also improve reliability.