Neural networks for analysing sports games
Deutscher übersetzter Titel: | Neuronale Netze zur Analyse von Sportspielen |
---|---|
Autor: | Perl, Jürgen; Tilp, Markus; Baca, Arnold; Memmert, Daniel |
Erschienen in: | Routledge handbook of sports performance analysis |
Veröffentlicht: | London: Routledge (Verlag), Taylor & Francis (Verlag), 2013, S. 237-247, Lit. |
Format: | Literatur (SPOLIT) |
Publikationstyp: | Sammelwerksbeitrag |
Medienart: | Elektronische Ressource (online) Gedruckte Ressource |
Sprache: | Englisch |
Schlagworte: | |
Online Zugang: | |
Erfassungsnummer: | PU201405004376 |
Quelle: | BISp |
Abstract des Autors
Two interlinked challenging tasks characterize the problem of analysing sports games: recording complex process game data and transforming them into useful information. These days, a significant part of the first task can be carried out using automatic position recording, thus placing increased emphasis on the second task; tracking a soccer game at 25 frames per second results in about 135,000 frames per game, which sums to about 6 million x-y coordinate data per game. Nevertheless, an experienced coach can filter significant information from these data and recognize patterns of player constellations in the playing processes. Neural networks using self-organizing maps (SOMs; see Kohonen, 1995) can recognize patterns in large data sets too and hence net-based data analysis can support the coach’s work. The first ideas of net-based analysis of sports games date back a few years, with recent advances in automatic position recording lending increased attention to complex games like soccer, handball, or basketball. Following a brief introduction to net-based game analysis, the chapter introduces the basics of net-based handling of process data and reports three instances detailing conceptual and methodical approaches of current research projects investigating handball, basketball, and soccer, respectively. Verf.-Referat