Technique models are used in various sports for technique training and for learning specific target techniques. They represent optimal, person-independent movement sequences that should enable athletes to solve a sporting task. The derivation of such technique models is mostly done theoretically (qualitatively) on the basis of biomechanical findings on technique execution.

The aim of this service research project in the work area Training & Movement funded by the Federal Institute of Sport Science (BISp) in cooperation with the Chair of Machine Learning and Machine Vision at the University of Augsburg was to develop empirically(quantitatively) derived technique models for the sport of swimming and to use them for performance diagnostics and prognosis.

Trained neural networks were used to determine the swimming technique of top athletes, since motion analysis with reflective markers at prominent body points is not an option in performance diagnostics for top swimmers.

The detected joint positions were then analyzed for common features using a factor analytic procedure (Principal Component Analysis, PCA). Based on the movement patterns obtained from the PCA, a mathematical-statistical description of a technique guideline for the individual swimming styles is provided.

https://www.youtube.com/watch?v=ggOKDxc-_Jk