Computer Science

News and events

30 April 2012

Algorithms identifying tactical variables in automatic tracking data in professional soccer

Presented By Dr Peter O'Donoghue, Reader, Cardiff Metropolitan University & Chair, International Society of Performance Analysis of Sport
  • 2.00pm
  • N112, Haslegrave Building

About this event

Abstract: Image processing has been utilised within advanced soccer analysis systems to provide player movement data relating to physiological demands and tactical aspects of play. There are various aspects of play that can potentially be analysed efficiently using player displacement data. These include support, depth, width and penetration in attack (Daniel, 2003), penetration in attack (Hargreaves and Bate, 2010), different types of support run performed during possession (Hughes, 1998, pressure, cover, back up and balance in defence (Tenga, 2010), compactness of defence (Daniel, 2003) and team placement during zonal and man to man defending (Prestigiacomo, 2003). All of these areas can potentially be recognised through automated analysis of player displacement data. For an algorithm to be developed to automatically analyse any aspect of play, there must be (a) an agreed understanding of tactical aspects and their characteristics at a subjective level, (b) precise definitions of location and movement patterns that characterise different tactical aspects and (c) a mathematical representation of the tactical aspects of interest. This is illustrated in the current research using two examples; (a) balance of defence (Olsen, 1981) and creating space in attack / denying space in defence (Bangsbo and Peitersen, 2004; Bangsbo and Peitersen, 2002). Olsen (1981) described balanced and unbalanced defences and there is evidence that attacking is more productive against unbalanced defences in international football (Olsen and Larsen, 1997). Tenga (2009) produced operational definitions for the three key aspects of balance of defence; pressure, back up and cover. An algorithm has been developed in Matlab version (The Matworks Inc., Natick, MA) to determine whether a defence is balanced at any point during an opposition attack using Tenga's (2009) definitions. A further algorithm has been developed to determine mean distance variables from nearest opponents for the team in possession. The algorithm do require a second data source to identify times of starts and ends of possessions as well as their classified outcomes. This was done for the forward most 1 to 10 players in the team for the first and last 3s, 4s and 5s of possessions as well as the difference in mean space between the beginning and the end of the possessions. The algorthm was tested for a Bundesliga match using one 45 minute half of football. The most significant variable was the mean space for the 10 outfield players during the final 5s of the possession (p <0.05). The mean distance to the nearest opponent was 7.5+1.4m during the last 5s when the 2 goals were scored 6.4+0.8m when there was a shot on target and 6.1+0.2m, when there was a shot off target. This variable may be a useful indicator for one team's ability to create space when in possession and another team's ability to restrict space when defending.