Action sequence analysis in team handball
The analysis of game situations in sports games is essential for development of successful game tactics and planning of training. Carling (2008) suggested analyzing action sequences because the study of single actions only gives restricted insight into team’s behavior. The aim of the present study i...
Main Authors: | , |
---|---|
Format: | Article |
Language: | English |
Published: |
University of Alicante
2013-09-01
|
Series: | Journal of Human Sport and Exercise |
Subjects: | |
Online Access: | http://www.jhse.ua.es/jhse/article/view/574 |
id |
doaj-d48e747672224f0089406a1d3383d143 |
---|---|
record_format |
Article |
spelling |
doaj-d48e747672224f0089406a1d3383d1432020-11-24T22:00:05ZengUniversity of AlicanteJournal of Human Sport and Exercise1988-52022013-09-0183Proc61562110.4100/jhse.2013.8.Proc3.07Action sequence analysis in team handballNorbert SchrapfMarcus TilpThe analysis of game situations in sports games is essential for development of successful game tactics and planning of training. Carling (2008) suggested analyzing action sequences because the study of single actions only gives restricted insight into team’s behavior. The aim of the present study is to analyze action sequences in team handball to identify offensive behaviors. For the study 6 games from the EURO-Men-18 in Austria were recorded. Special categories for annotation were defined to assess single actions which then have been merged into action sequences. Shots and up to 5 passes prior the shot were annotated with custom-made software. Out of 3212 actions, each containing information about video time stamp and ground position, the software generated 612 action sequences. To identify different behaviours, similar action sequences were determined using artificial neuronal network software (Perl, 2002). To optimize network performance the dataset was enlarged with noise of 15% to a quantity of 3060 action sequences. Subsequently, the network with a dimension of 400 neurons was trained. Each neuron represents an action sequence pattern. Similar neurons are grouped to clusters representing similar behaviour. The artificial network recognized 32 clusters. Additional, 10 single neurons could not be classified to a cluster. Therefore, 42 different offensive team behaviours were identified whereby 8 clusters represented 49% of the actions sequences. The study revealed the potential to identify playing patterns by analyzing action sequences with artificial neuronal networks. Expert review of the recognized patterns showed a promising accordance with actual playing patterns. Future steps will be the detection of preferred tactics in single teams, the integration of goal success and the identification of successful offensive tactics.http://www.jhse.ua.es/jhse/article/view/574TEAM HANDBALLARTIFICIAL NEURONAL NETWORKOFFENSIVE PATTERNSTEAM SPORTS |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Norbert Schrapf Marcus Tilp |
spellingShingle |
Norbert Schrapf Marcus Tilp Action sequence analysis in team handball Journal of Human Sport and Exercise TEAM HANDBALL ARTIFICIAL NEURONAL NETWORK OFFENSIVE PATTERNS TEAM SPORTS |
author_facet |
Norbert Schrapf Marcus Tilp |
author_sort |
Norbert Schrapf |
title |
Action sequence analysis in team handball |
title_short |
Action sequence analysis in team handball |
title_full |
Action sequence analysis in team handball |
title_fullStr |
Action sequence analysis in team handball |
title_full_unstemmed |
Action sequence analysis in team handball |
title_sort |
action sequence analysis in team handball |
publisher |
University of Alicante |
series |
Journal of Human Sport and Exercise |
issn |
1988-5202 |
publishDate |
2013-09-01 |
description |
The analysis of game situations in sports games is essential for development of successful game tactics and planning of training. Carling (2008) suggested analyzing action sequences because the study of single actions only gives restricted insight into team’s behavior. The aim of the present study is to analyze action sequences in team handball to identify offensive behaviors. For the study 6 games from the EURO-Men-18 in Austria were recorded. Special categories for annotation were defined to assess single actions which then have been merged into action sequences. Shots and up to 5 passes prior the shot were annotated with custom-made software. Out of 3212 actions, each containing information about video time stamp and ground position, the software generated 612 action sequences. To identify different behaviours, similar action sequences were determined using artificial neuronal network software (Perl, 2002). To optimize network performance the dataset was enlarged with noise of 15% to a quantity of 3060 action sequences. Subsequently, the network with a dimension of 400 neurons was trained. Each neuron represents an action sequence pattern. Similar neurons are grouped to clusters representing similar behaviour. The artificial network recognized 32 clusters. Additional, 10 single neurons could not be classified to a cluster. Therefore, 42 different offensive team behaviours were identified whereby 8 clusters represented 49% of the actions sequences. The study revealed the potential to identify playing patterns by analyzing action sequences with artificial neuronal networks. Expert review of the recognized patterns showed a promising accordance with actual playing patterns. Future steps will be the detection of preferred tactics in single teams, the integration of goal success and the identification of successful offensive tactics. |
topic |
TEAM HANDBALL ARTIFICIAL NEURONAL NETWORK OFFENSIVE PATTERNS TEAM SPORTS |
url |
http://www.jhse.ua.es/jhse/article/view/574 |
work_keys_str_mv |
AT norbertschrapf actionsequenceanalysisinteamhandball AT marcustilp actionsequenceanalysisinteamhandball |
_version_ |
1725845420840583168 |