Decoding speed climbing: AI-powered biomechanical analysis of elite performance

Abstract

This study explores the biomechanical factors that affect speed climbing performance using advanced AI-based motion analysis. Design: A retrospective analysis was performed using video recordings from 11 international competitions (2017–2023), focusing on three top athletes with the highest number of performances (n = 56). Methods: A computer vision system with a convolutional neural network enabled markerless motion analysis, to extract key kinematic parameters, including as center of gravity trajectory, contact times, limb frequency, and flow parameters. Multiple regression analyses with backward elimination identified significant predictors of End Time, using a significance threshold at p < 0.05. Results: Significant correlations were found between hand limb frequency measures and End Time across all athletes (r = 0.437–0.640, p < 0.001). Regression analysis revealed athlete-specific predictors: Limb Frequency was the strongest predictor for Athlete 1 (β = 0.900), while Athlete 2’s performance was influenced by overall limb frequency (β = -0.625) and maximum foot load (β = 0.638). Athlete 3 demonstrated an alternative strategy characterized by rapid foot transitions and consistent hand movement rhythm. Analysis of contact times showed that Athlete 2 had the shortest average hand contact time (0.35 ± 0.087 s), indicating superior efficiency in hand repositioning. Athlete 1 exhibited anomalies in foot contact parameters, including a notably long minimum foot contact time, potentially reflecting technical inefficiencies or variability in movement execution. Athlete 3 performed closest to the group average but showed greater consistency across trials. Multicollinearity issues highlighted the interdependence of biomechanical parameters, emphasizing the need for advanced analytical approaches. Conclusions: Speed climbing performance is characterized by distinct athlete-specific patterns, underscoring the importance of personalized training methodologies. The study demonstrates the potential of AI-driven systems for detailed biomechanical analysis, while also highlighting the need for future research incorporating larger datasets, longitudinal designs, and physiological measures to provide a more comprehensive understanding of speed climbing performance determinants.