In this paper we introduce two fundamentally different techniques for optimizing counter formations in team sports. In the first technique, we use canonical correlation analysis (CCA) to learn an "explicit" relationship between offensive and defensive formations. We then use the learned CCA components to make predictions about players' spatial position. Experimenting with the basketball dataset (NBA season 2012-2013) we are able to predict players' positions with high precision. In the second technique, we create an image-based representation of the player movements relative to the ball. The mentioned representation enables coaches to assess team formations in a glance. The recently developed Radon Cumulative Distribution Transform (RCDT) was used alongside CCA to analyze the image-based representations. With these techniques, we provide real-time feedback to optimize both players' positions and team formations.
© Copyright 2017 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE. Veröffentlicht von IEEE. Alle Rechte vorbehalten.