Pose estimation or manual digitising: Can automating technologies change the current in-field assessment of high jump?

Biomechanists spend significant time completing the time-consuming task of manually digitising 2D videos to derive kinematic spatiotemporal parameters. Recent advances in 2D pose estimation models (PEMs) hold promise for automating the determination of parameters in sport. This study developed an automated PEM digitising and analysis pipeline (AAP) for high jump. We investigated differences in four spatiotemporal and joint angle outputs from traditional manual processing pipelines (MAP) and the AAP using paired t-tests, intra-class correlations and effect size analysis. Statistical analysis revealed that knee angles derived from the MAP and AAP were not different, whereas penultimate foot contact time and both body angle “lean” measures were different. The custom AAP considerably reduced processing time for the selected high jump execution parameters.