BACKGROUND: Markerless motion analysis based on computer vision has expanded in sport sciences, but its validity in aquatic environments remains unclear due to optical distortions and fluid-related constraints. This study aimed to evaluate the feasibility, precision and validity of an off-the-shelf pose estimation tool for underwater kinematic analysis using a controlled cyclic movement.; METHODS: Eight physically active participants performed aquatic cycling at 40, 60 and 80 rpm. Hip, knee and ankle anatomical landmarks were identified using a manual tracking approach (Kinovea) and a markerless algorithm (Sports2D). Spatial accuracy was assessed through root mean square error (RMSE) of landmark detection, while knee joint angle agreement was evaluated using correlation analysis and Bland-Altman methods; RESULTS: RMSE values remained stable across cadences (p > 0.05), with lower errors observed at the knee compared to hip and ankle (p < 0.01). A strong linear relationship between methods was found for knee joint angles (r² = 0.67-0.88), although systematic differences were present. Sports2D underestimated minimum angles and overestimated maximum angles, indicating an amplification of movement amplitude and limits of agreement remained consistent across cadences; CONCLUSIONS: The markerless approach demonstrated consistent anatomical landmark detection in an aquatic environment, but showed systematic bias in joint angle estimation. These findings support its use for applications requiring reliable landmark tracking, while highlighting the need for caution and context-specific validation when interpreting joint kinematics.
© Copyright 2026 XVth International Symposium on Biomechanics and Medicine in Swimming Proceedings. Alle Rechte vorbehalten.