Last modified: 2026-05-31
Abstract
Subjective coaching in running lacks quantitative precision, while optical motion capture systems is too restrictive for daily track environments. This study proposes a markerless proficiency evaluation framework using high-speed cameras for data-driven coaching.
Deep learning-based pose estimation has shown great potential in sports biomechanics. In this study, AlphaPose extracts 2D skeletal coordinates, focusing on swing-phase knee flexion—a critical indicator of running efficiency. To address cadence variations, Dynamic Time Warping (DTW) and DTW Barycenter Averaging (DBA) are applied to expert kinematic data. While DTW effectively synchronizes athletic motion, DBA generates an optimal reference template without extensive labeled datasets.
Athlete proficiency is evaluated by calculating the DTW distance between each athlete's waveforms and the expert template. By converting subjective coaching intuition into objective scores, this system identifies gait inefficiencies, facilitating personalized training and reducing injury risks in collegiate athletes