Evaluation of a smartphone-based markerless motion capture system for assessing the kinematics of gymnastics

Monitoring gymnasts in their training environments, using markerless motion capture, could enhance our understanding of their injuries and facilitate the implementation and evaluation of injury prevention measures. This study compared a smartphone-based markerless motion capture system (OpenCap) to a gold standard marker-based system (Vicon) to see how well it could track basic gymnastics. Ten participants were recorded simultaneously using both systems whilst performing: walk, squat, sit-to-stand, drop jump, handstand, cartwheel, handstand walk, handstand hop. Comparisons were made between gymnastics and non-gymnastics movements, and the effects of musculoskeletal range of motion (ROM) constraints and body region (upper vs lower body) were evaluated. System performance was assessed using complimentary metrics: magnitude of error between systems, waveform similarity, agreement and bias, and the relative magnitude of inter-system error compared to inter-subject variability. For gymnastics movements, using a musculoskeletal model with ROM constraints adapted to ranges typically utilised significantly reduced root mean square error (RMSE) between Vicon and OpenCap. However, the mean RMSE across all joints for gymnastics movements (40.14° ± 14.48°) was still significantly higher than non-gymnastics movements (11.57° ± 6.66°). Kinematic errors were lower for lower- than upper- limb joints, with the greatest discrepancies observed at the pelvis and shoulder during gymnastics movements. Although movement-specific modelling approaches can improve performance, OpenCap currently tracks complex gymnastics movements less accurately than simpler functional tasks. Further developments, including training on a wider range of activities and implementation of appropriate movement-specific constraints, are required before markerless motion capture can be applied with confidence to gymnastics.