This article was downloaded by: [University of Chicago Library]On: 25 September 2013, At: 14:26Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Sports BiomechanicsPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/rspb20

Quantitative analysis of Brazilianfootball players' organisation on thepitchFelipe Arruda Moura a , Luiz Eduardo Barreto Martins b , RicardoDe Oliveira Anido c , Ricardo Machado Leite De Barros a & SergioAugusto Cunha aa Laboratory of Instrumentation for Biomechanics, College ofPhysical Education, Campinas State University, Campinas, Brazilb Laboratory of Instrumentation for Physiology, College of PhysicalEducation, Campinas State University, Campinas, Brazilc Institute of Computing, Campinas State University, Campinas,BrazilPublished online: 05 Dec 2011.

To cite this article: Felipe Arruda Moura , Luiz Eduardo Barreto Martins , Ricardo De OliveiraAnido , Ricardo Machado Leite De Barros & Sergio Augusto Cunha (2012) Quantitative analysisof Brazilian football players' organisation on the pitch, Sports Biomechanics, 11:1, 85-96, DOI:10.1080/14763141.2011.637123

To link to this article: http://dx.doi.org/10.1080/14763141.2011.637123

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,

systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

Quantitative analysis of Brazilian football players’organisation on the pitch

FELIPE ARRUDA MOURA1, LUIZ EDUARDO BARRETO MARTINS2,

RICARDO DE OLIVEIRA ANIDO3, RICARDO MACHADO LEITE DE

BARROS1, & SERGIO AUGUSTO CUNHA1

1Laboratory of Instrumentation for Biomechanics, College of Physical Education, Campinas State

University, Campinas, Brazil, 2Laboratory of Instrumentation for Physiology, College of Physical

Education, Campinas State University, Campinas, Brazil, and 3Institute of Computing, Campinas

State University, Campinas, Brazil

(Received 19 October 2010; accepted 14 August 2011)

AbstractThe purpose of this study was to characterise Brazilian teams’ coverage area and spread on the pitchwhile attacking and defending and to analyse the teams’ organisation in tackle and shot on goalsituations. We obtained the trajectories of 223 players in eight games with a tracking method. Team areawas defined as the area of the convex hull formed by players’ positions. Team spread was defined as theFrobenius norm of the distance-between-player matrix. We calculated teams’ area and spread over timeand in situations of shots on goal (n ¼ 233) and tackles (n ¼ 1,897). While the players attacked, spreadand area (median ^ confidence interval) ranged from 322.9 ^ 0.8 to 387.8 ^ 1.0 m and from905.4 ^ 4.4 to 1,407.6 ^ 5.5 m2, respectively. On defence, the values were smaller (p , 0.05) andranged from 283.4 ^ 0.9 to 325.8 ^ 0.9 m and from 773.8 ^ 4.6 to 1,158.4 ^ 5.5 m2 for the spreadand the area. In defending circumstances, the teams presented a greater area and spread when theysuffered shots on goal than when the teams performed tackles. In attacking situations, the teamspresented a greater area and spread when they suffered tackles than when they performed shots on goal.The results allowed showing the attacking–defending interaction between Brazilian teams.

Keywords: Computational tracking, football, tactics, team coverage area, spread

Introduction

An objective analysis of a sport requires developing methodologies for gathering information

about the players’ actions during the competition to examine the physical, technical, and

tactical features of the game. Technology enhancements have led to the development of

player-tracking systems that supply player trajectory data (Carling et al., 2008). These

systems are widely used to quantify the physical demands of football players in multiple

countries (Barros et al., 2007; Di Salvo et al., 2007; Bradley et al., 2009; Di Salvo et al.,

2009).

ISSN 1476-3141 print/ISSN 1752-6116 online q 2012 Taylor & Francis

http://dx.doi.org/10.1080/14763141.2011.637123

Correspondence: Felipe Arruda Moura, Laboratory of Instrumentation for Biomechanics, College of Physical Education, Campinas

State University, Rua Manuel Jose Machado, 67 apto. 53, CEP 04676-100 Sao Paulo, Brazil, E-mail: felipeunesp@yahoo.com.br

Sports Biomechanics

March 2012; 11(1): 85–96

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

Compared to the number of studies in the literature which deals with the physical efforts of

football players, research on tactical analysis is limited. A frequent football tactic is that when

a team attacks, players have to solve the complex problems of how to maintain possession of

the ball, attack the goal, and move to empty regions of the pitch to create scoring

opportunities. However, when the team is defending, the players should move to protect

their goal and to win back possession (Mitchell, 1996). Therefore, the organisation of the

players on the pitch can explain important game tactics, and ball possession may change

through that organisation.

Team coverage area and spread on the pitch are pertinent variables to understanding the

players’ organisation during a football game, and they may represent tactical performance

indicators (Garganta, 1997; Hughes & Bartlett, 2002; Frencken & Lemmink, 2009).

Nevertheless, only three studies were found in the literature which relate to the analyses

with these quantitative data. The team coverage area was analysed by Okihara et al. (2004)

in two Japanese football matches. However, the authors defined the team area using the

position information of only four players. A recent study also analysed team coverage area

during attacks (Frencken & Lemmink, 2009), but this analysis was applied to two four-a-

side games, which may not represent the actual situation of a high-level competitive

football match. Another study defined the instantaneous radius of each team as the

average distance between all players and the geometrical centre of the team to quantify

their tactical characteristics when the teams were with and without ball possession (Yue

et al., 2008a). The authors presented results from only one half of a game between

German teams.

Therefore, studies analysing football team organisation on the pitch have used small sample

sizes, and no studies were found to use Brazilian teams. Based on the data about the frequency

of passes, crosses, shots to goal, and other technical actions, a previous investigation

(Yamanaka et al., 1993) showed that South American teams have a unique pattern of play

compared with other countries. However, no data were provided about the players’

organisation on the pitch.

During specific situations of a game, team organisation may change as a result of a

perturbation such as a loss of possession or a scoring of a goal (Frencken & Lemmink, 2009).

Therefore, an analysis of team coverage area and spread on the pitch in tackle and shot on

goal situations can provide valuable tactical information. It can be expected that when teams

perform a tackle while defending, they have a different organisation than when they suffer a

shot on their goal. However, teams may have a different organisation when performing an

offensive shot on the opponent’s goal than when they suffer a tackle.

Once the player-tracking systems register the players’ positions, a measure of team

coverage area and spread on the pitch can be extracted. For a better understanding of

football dynamics, it is necessary to comprehend how these variables change during the game

and when teams are attacking or defending. Additionally, a specific analysis of a team’s

coverage area and spread on the pitch in situations of tackles and shots to goal can provide

valuable insights for the coaches to improve training programmes and thus competitive team

performance. Therefore, the purposes of this study were (a) to characterise the Brazilian

teams’ coverage area and spread on the pitch while attacking and defending and (b) to

analyse the teams’ organisation in tackle and shot on goal situations. Specifically, we were

interested in evaluating whether teams while defending have a different organisation when

they perform a tackle than when they suffer a shot on goal. Also, we analysed whether teams

in an attacking situation have a different organisation on the field when performing shots to

goal than when suffering tackles.

F. A. Moura et al.86

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

Methods

Data collection

The Ethics Committee of the Sao Paulo State University approved this research. At least four

digital cameras (30 Hz) filmed eight Brazilian First Division Championship matches

between 16 different teams. To facilitate identification, teams were labelled tm1, tm2, . . . ,

tm16. The cameras were fixed at the stadiums’ highest points, each covering roughly a

quarter of the pitch while including overlapping regions. After the games, we transferred the

video sequences to our computers for analysis, and then we synchronised the cameras,

identifying the common events occurring in the overlapping regions, such as a player’s kick.

Participants and tracking method

Using an automatic tracking method via a DVideo software interface (Barros et al., 2006;

Figueroa et al., 2006a, 2006b), we obtained the trajectories of 223 football players in the

eight games. DVideo software has an automatic tracking rate of 94% for the processed frame,

an average error of 0.3 m for the player position determination, and an average error of 1.4%

for the distance covered by players (Figueroa et al., 2006b; Misuta, 2009).

Prior to the games, we measured specific points on the pitch with a tape measure to

calculate image–object transformations for the calibration of the cameras. After measuring

the players’ positions from the video sequences, we constructed 2D coordinates on a pitch

coordinate system using a direct linear transformation (DLT; Abdel-Aziz & Karara, 1971).

Figure 1 shows the pitch coordinate system (X, Y).

We numbered players on each team p ¼ 1, 2, . . . , 11. Thus, for each team analysed, the

2D coordinates of the players are defined as follows:

XpðtÞ;YpðtÞjt ¼1

f;2

f; . . . ;

N

f

� �; ð1Þ

Figure 1. Pitch coordinate system.

Football players’ organisation on the pitch 87

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

where t represents each instant of time (in seconds), f represents the frequency of the video

sequence, and N represents the number of video sequence frames.

A Butterworth third-order low-pass digital filter with a cut-off frequency of 0.4 Hz filtered

the 2D coordinates of all the players. The choice of the cut-off frequency was based on the

two protocols described in Misuta (2004). The first protocol consisted of a dynamic test. In

this test, we asked a participant to cover a known distance walking, jogging, and sprinting.

We applied the same tracking procedures, and then we filtered the 2D coordinates with the

Butterworth low-pass filter with different orders and cut-off frequencies. After each filtering

trial, we calculated the distance covered by the participant and compared it with the real

distance. The third order with a cut-off frequency of 0.4 Hz was the combination that

presented the best results. In the second procedure, we performed a residual analysis (Cunha

& Lima Filho, 2003) that confirmed these parameters as good choices.

Having obtained the trajectories of the players for the whole game, we were able to

calculate the teams’ coverage area and spread.

Ball possession

After obtaining the players’ 2D coordinates, DVideo software was used to determine the

players’ technical actions. DVideo software has an interface developed for studying football

matches. While the operator watches the match in DVideo software, when an event happens

(passing, shot on goal, tackle, foul, etc.), he identifies with the mouse in a bar which action

was performed and which player performed this action. Once the players’ 2D coordinates as

a function of time were determined, at the end of the analysis, we created a matrix that stored

the technical action information of all the players, the instant when these actions occurred,

and the 2D coordinates of where the players were positioned during the action. With this

information, an algorithm was created to identify when teams were with or without ball

possession and the instances of tackles and shots to goal. In this study, we considered that a

team recovered ball possession when any player(s) performed two consecutive technical

actions (i.e. two actions performed by a single player or one action performed by two

different players on the same team). Thus, according to this criterion, a team did not lose ball

possession when the opposing team performed a foul or unsuccessfully attempted to tackle.

Moreover, ball possession was attributed to the team that performed the next technical

action when the ball was out of play. We adopted this criterion because even when the ball

was out of play, the teams organised themselves based on whether they had ball possession in

the next action.

With this information, we analysed the coverage area and the players’ spread on the pitch

when teams were attacking or defending (in other words, when teams were with or without

ball possession, respectively) and in shot on goal and tackle situations. To reduce the amount

of data to be processed, for all games analysed, the teams’ coverage area and spread were

calculated at a frequency of 7.5 Hz.

Team spread

For each instant of time t, we calculated the Euclidian distance between each player and his

teammates. The distances between players were organised in a symmetric matrixD(t) of order

11 (11 players on a team). Because D is a symmetric matrix, dij equals dji; in other words, the

element d12 is equal to d21, and it represents the Euclidian distance between p ¼ 2 and 1. In

addition, the principal diagonal is null because the Euclidian distance between a player and

himself is zero. Therefore, we considered only the values of the lower triangular matrix L.

F. A. Moura et al.88

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

The Frobenius norm is one of the most frequently used matrix norms in linear algebra and

provides a measure of distance (Golub & Van Loan, 1989). Thus, we calculated the

Frobenius norm for matrix L (designated kLkF) for each instant of time (t) to represent the

team spread:

kLðtÞkF ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXni¼1

Xnj¼1

jlij j2

vuut : ð2Þ

Thus, according to Equation (2), large values of the Frobenius norm of the matrix L

characterise players spread across the football pitch, whereas low values characterise players

close to each other.

Team coverage area

The area that a team covers is also related to the players’ pitch position. One way to

represent this is via the convex hull area. The convex hull of a set of points S on a plane (in

our case, represented by each player’s position on the same team in each instant of time t)

is the smallest convex set containing S; if S is finite, the convex hull is always a polygon

whose vertices are a subset of S (Preparata & Shamos, 1985). The convex hull was

computed by the Quickhull technique (Barber et al., 1996), which is available in the

Matlabw software.

At each instant of time t, we divided the team convex hull into triangles to aid the

calculation of the convex hull area CA(t) (i.e. we summed the areas of all the triangles within

the convex hull). Figure 2 shows an example of a convex hull, the triangulation, and the

convex hull areas for both teams.

Shot on goal and tackle situations

We analysed the teams’ coverage area and spread for 233 specific situations of shots on goal

and 1,897 tackle situations. In a defending situation, the coverage area and spread were

identified in the exact frame when the team performed a tackle or suffered a shot on their

goal. Equally, in an attacking situation, these variables were analysed in the exact frame when

the team performed a shot on goal or suffered a tackle.

Statistical analyses

Prior to each analysis, a Lilliefors test verified whether the data were normally distributed.

We adopted a , 0.05 for all statistical analyses. All distributions were abnormal ( p , 0.01);

thus, we used non-parametric tests in the following statistical analyses.

A Wilcoxon rank-sum test compared the team spread and coverage area for when the

teams had ball possession versus when they did not. These values are expressed as

medians ^ confidence intervals (CIs) and were calculated based on recommendations from

McGill and colleagues (1978).

Spearman’s rank-order correlation tested the relationship between the Frobenius norm of

distance-between-player matrices and the convex hull areas for each match.

The team coverage area and spread were analysed in specific shot on goal and tackle

situations. Thus, we were interested in evaluating whether, in a defending circumstance,

teams presented a different organisation when tackles were performed than when they

Football players’ organisation on the pitch 89

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

suffered shots on their goal. Also, we analysed whether, in an attacking circumstance, teams

presented a different coverage area and spread when they performed shots on goal than

when they suffered tackles. In both comparisons, a Wilcoxon rank-sum test was applied

and the values are also expressed as medians ^ CIs.

Results

Figure 3 shows examples of coverage area (A) and spread (B) time-series changes from

both tm1 and tm2 during 10 min of the first half of the game. Visually, the teams’ spread

time series show a counter-phase relation, but their coverage area time-series do not.

This behaviour was found throughout the duration of the game for all the matches

analysed.

Additionally, Figure 4 shows an example of the coverage area (A) and spread (B) time-

series during a ball possession exchange. It is possible to verify that when a team loses ball

possession, the coverage area, and the spread decrease. However, when a team wins ball

possession, these variables increase.

Table I confirms the results of Figure 4 for all the studied teams for the duration of the

games. The Wilcoxon rank-sum test showed that all the teams had greater values of spread

and coverage area when they had the ball compared to when they did not.

A Spearman rank-order correlation coefficient (r) analysed the relationship between the

Frobenius norm of the distance-between-player matrices and the convex hull areas (which

70A B

60

50

40

30Y (

m)

20

10

00 10 20 30 40

x (m)50 60 70 80 90 100

70

60

50

40

30Y (

m)

20

10

00 10 20 30 40

x (m)50 60 70 80 90 100

70C

60

50

40

30Y (

m)

20

10

00 10 20 30 40

x (m)50 60 70 80 90 100

Figure 2. Player position and team convex hull (A), convex hull triangulations (B), and convex hull areas (C).

F. A. Moura et al.90

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

represents team spread and coverage area, respectively) for each match (Table II).

Spearman’s test confirmed that there was a positive correlation between team coverage area

and spread variable for all teams.

We calculated the teams’ coverage area in a defending situation when teams performed

tackles or suffered shots on goal. Also, these variables were calculated in an attacking

situation, when teams performed shots on goal or suffered tackles. Table III shows that, in

defending circumstances, teams presented greater coverage area and spread ( p , 0.01)

when they suffered shots to goal, compared to when teams succeeded in performing tackles.

In attacking situations, teams presented greater coverage area and spread when suffering

tackles, compared to when they succeeded in performing shots on goal.

Discussion

In a football match, the players assume their positions on the pitch according to the dynamics

of the game as a function of time. In this study, two variables described the players’

organisation to characterise the Brazilian teams’ tactics. The Frobenius norms of the

distance-between-player matrices and the convex hull areas represented team spread and

coverage area, respectively. Ball possession exchanges caused changes in the players’

distribution on the pitch during the match.

The results confirm that when defending, Brazilian football players move closer to each

other (as shown by the Frobenius norms) and reduce the pitch coverage area (as shown by

convex hull areas). Conversely, when the teams were attacking, these variables increased.

Figure 3. Teams’ coverage area (A) and spread (B) during 10 min of the first half of the game.

Football players’ organisation on the pitch 91

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

Okihara et al. (2004) reported the same team behaviour in two Japanese football matches.

In that case, the authors determined only the teams’ coverage area using the position of four

players, whereas in the present study, we used the convex hull area, allowing us to take the

position of all the players (except the goalkeeper) into consideration. The authors showed that

without ball possession, the coverage area of Japanese teams ranged from 1,207.1 ^ 631.8 to

1,392.4 ^ 460.9 m2. With ball possession, Japanese teams areas ranged from 1,379.6 ^ 570

to 1,703 ^ 569.2 m2. The differences between area calculations may explain why the values

presented by Okihara et al. were mostly larger than those in our research.

In the present research, we observed that the players were more distant from each other

when the team possessed the ball and were closer when the team did not, according to the

teams’ spread values. Also, a visual analysis of the teams’ spread temporal-series presented a

counter-phase relation: when the team spread increases, the opponent’s team spread

decreases, and vice versa. These data corroborate the temporal-series of two team geometric

radii presented by Yue et al. (2008a). However, the coverage area temporal-series did not

present a clear counter-phase relation, showing that the team spread and the coverage area

provide different information about the players’ distribution on the field.

As we presented in the results, there was a positive correlation between team coverage area

and spread. However, eventually, a team may present elevated convex hull areas, but this is

not associated with large spread values. Likewise, the players’ organisation may provide a

small coverage area but not low spread values. These circumstances can be explained by the

fact that one player who is far away from his teammates yields a large convex hull area, but

Figure 4. Teams’ coverage area (A) and spread (B) during a ball possession exchange.

F. A. Moura et al.92

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

Table II. The relationship between team coverage area and team spread.

Team r Team r

Match 1 tm1 0.60* tm2 0.77*

Match 2 tm3 0.66* tm4 0.75*

Match 3 tm5 0.52* tm6 0.82*

Match 4 tm7 0.77* tm8 0.81*

Match 5 tm9 0.70* tm10 0.66*

Match 6 tm11 0.69* tm12 0.75*

Match 7 tm13 0.79* tm14 0.57*

Match 8 tm15 0.66* tm16 0.80*

*p , 0.05.

Table III. Teams’ coverage area and spread (median ^ CI) in the specific situations of shots on goal (n ¼ 233) and

tackles (n ¼ 1897).

Defending situation Attacking situation

Variables When tackles

were performed

When teams suffered

shots to goal

When teams

suffered tackles

When shots to goal

were performed

Coverage area (m2) 920.7 ^ 13.3* 1,110.4 ^ 41.7 1,059.6 ^ 15.2* 898.9 ^ 43.9

Team spread (m) 304.9 ^ 2.4* 393.7 ^ 5.5 349.8 ^ 3.1* 277.2 ^ 7.6

*p , 0.05.

Table I. Median and CI values of team spread (m) and coverage area (m2), with and without ball possession.

Team spread (m) Team coverage area (m2)

With ball

possession

Without ball

possession

With ball

possession

Without ball

possession

Teams Median CI Median CI Median CI Median CI

Tm1 367.0* 0.7 325.8 0.9 1,114.0* 4.0 961.5 4.2

Tm2 338.5* 0.9 304.7 0.7 1,095.3* 6.1 943.2 3.6

Tm3 354.9* 0.6 316.9 0.8 1,127.4* 4.3 1,009.3 4.0

Tm4 340.1* 0.8 298.7 0.8 1,068.2* 4.8 878.1 3.9

Tm5 387.8* 1.0 315.5 0.8 1,407.6* 5.5 1,158.4 5.5

Tm6 370.1* 1.1 312.3 0.6 1,257.3* 5.4 1,015.4 4.0

Tm7 360.8* 0.8 317.3 0.9 1,210.8* 5.0 990.3 5.1

Tm8 343.6* 0.6 318.6 0.7 1,169.3* 4.9 1,032.9 4.4

Tm9 338.7* 0.7 301.3 0.6 985.3* 4.4 867.6 3.3

Tm10 333.2* 0.8 295.2 0.8 967.6* 4.0 813.9 3.6

Tm11 353.0* 0.7 304.6 0.8 1,031.3* 4.4 862.1 3.6

Tm12 327.3* 0.7 287.4 0.7 976.2* 4.8 809.9 4.3

Tm13 350.0* 0.7 295.1 0.7 1,033.2* 4.5 805.5 3.4

Tm14 335.7* 0.6 295.1 0.7 986.5* 3.8 845.2 3.5

Tm15 350.3* 0.7 306.6 0.8 976.2* 4.0 858.5 3.9

Tm16 322.9* 0.8 283.4 0.9 905.4* 4.4 773.8 4.6

*p , 0.05, greater than ‘without ball possession’ group.

Football players’ organisation on the pitch 93

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

once all the other players are grouped, the Frobenius norm is reduced. Another exception

happens if all the players are positioned in a wall during an adversary free kick; in this

situation, the team coverage area is very small but not the spread measure. These examples

may explain why some correlations coefficients were below 0.6 and why coverage area and

spread temporal-series exhibit different behaviours. Therefore, team coverage area and

spread provide particular information about the players’ distribution on the pitch, and

together, they illuminate the game’s dynamics.

Football is complex, and even a great quantity of measured variables may not explain or

determine match scores. For this reason, the literature affirms that football matches are

stochastic processes (Yue et al., 2008a, 2008b). However, player movements and organisation

on the pitch have deterministic features that explain important characteristics of the game, as

this study has verified.

During the training process, it is very usual that coaches instruct their players to organise

the positions on the field according to their tactical choices. For example, a team can

compact its players every time the opponent passes beyond the midline with ball possession.

Additionally, when attacking, a team can also organise its players according to the playing

patterns established by the coach. Castelo (1999) suggests that in elaborated attacks, teams

present an organisation that evidences a compact structure. However, this behaviour was not

found in the Brazilian teams analysed in the present study. Therefore, further analyses of

teams from different countries can help researchers to better understand and identify various

playing patterns.

When defending, some of the team’s purposes are to avoid suffering a shot on their goal

and to recover ball possession. Inversely, when attacking, the players organise themselves to

increase shot on goal opportunities and to avoid being tackled (Mitchell, 1996). Therefore,

in the analysis of shot on goal and tackle situations, we were interested in evaluating whether

Brazilian teams have a different defensive organisation on the pitch that may assist them in

performing tackles and that may prevent shots on their goals. Additionally, we analysed

whether Brazilian teams have a different offensive organisation on the pitch that assists their

success in performing shots on goal and that prevents their players from being tackled.

The results show that in a defending situation, the teams had lower values of area and

spread when tackles were performed than when the teams suffered shots to their goals. These

data may indicate that teams are more likely to suffer shots on goal when they are not

compacted enough. While attacking, the teams presented a lower coverage area and spread

when performing shots on goal, compared to when they suffered tackles. Therefore, these

results can help coaches increase their teams’ chances of recovering ball possession while

defending and of performing shots on goal while attacking.

The method and results presented in this study are valuable tools to help coaches to

improve training and team performance. Our analyses contribute to understanding football

dynamics and provide important data about the features of the tactical actions of Brazilian

teams. Automatic tracking methods during training sessions allow coaches to calculate the

same variables proposed in the present research, and based on this information, they can

precisely control their players’ organisation on the pitch and systematise tactical strategies.

Additionally, these variables can be analysed after the game, and corrections can be

performed if necessary.

Conclusions

The purpose of this study was to characterise the organisation of Brazilian football players on

the pitch using team spread and coverage area measures and to analyse these variables in the

F. A. Moura et al.94

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

specific situations of shots on goal and tackles. We confirmed that when the teams had ball

possession, the players were more distant from each other and covered a greater area. When

the teams did not have the ball possession, the players were more compact and controlled a

smaller pitch area. In addition, the teams presented a lower coverage area and spread when

tackles were performed compared with when they suffered shots on goal. When attacking,

the teams’ coverage area and spread were lower when taking shots on the opponent’s goal,

compared with when they suffered tackles. With these results, we are able to show the

attacking–defending interaction between Brazilian teams and how players organise

themselves during the game.

Future studies should to attempt to identify other geometrical information such as the

intersection area between teams, which can provide data regarding team marking and area

dominance in relation to its opponent. Furthermore, the spread variable can be calculated

between specific players (e.g. among defenders and midfielders) to correlate their

organisation on the pitch with their roles in the match.

Acknowledgements

The authors would like to thank FAPESP, CAPES, CNPq, and Rede Globo de Televisao.

References

Abdel-Aziz, Y. I., & Karara, H. M. (1971). Direct linear transformation from comparator coordinates into object space

coordinates in close-range photogrammetry. Paper presented at the Proceedings of the Symposium on Close-Range

Photogrammetry Illinois.

Barber, C. B., Dobkin, D. P., & Huhdanpaa, H. (1996). The Quickhull algorithm for convex hulls. ACM

Transactions on Mathematical Software, 22 (4), 469–483.

Barros, R. M. L., Russomanno, T. G., Brenzikofer, R., & Figueroa, P. J. (2006). A method to synchronise video

cameras using the audio band. Journal of Biomechanics, 39 (4), 776–780.

Barros, R. M. L., Misuta, M. S., Menezes, R. P., Figueroa, P. J., Moura, F. A., Cunha, S. A., . . . Leite, N. J. (2007).

Analysis of the distances covered by first division Brazilian soccer players obtained with an automatic tracking

method. Journal of Sports Science and Medicine, 6, 233–242.

Bradley, P. S., Sheldon, W., Wooster, B., Olsen, P., Boanas, P., & Krustrup, P. (2009). High-intensity running in

English FA Premier League soccer matches. Journal of Sports Science, 27 (2), 159–168.

Carling, C., Bloomfield, J., Nelsen, L., & Reilly, T. (2008). The role of motion analysis in elite soccer: Contemporary

performance measurement techniques and work rate data. Sports Medicine, 38 (10), 839–862.

Castelo, J. F. F. (1999). Futbol: estructura y dinamica del juego. Barcelona: INDE Publicaciones.

Cunha, S. A., & Lima Filho, E. C. (2003). Metodologia para suavizacao de dados biomecanicos por funcao nao

parametrica ponderada local robusta. Brazilian Journal of Biomechanics, 1 (6), 23–28, in Portuguese: English

abstract.

Di Salvo, V., Baron, R., Tschan, H., Calderon Montero, F. J., Bachl, N., & Pigozzi, F. (2007). Performance

characteristics according to playing position in elite soccer. International Journal of Sports Medicine, 28 (3),

222–227.

Di Salvo, V., Gregson, W., Atkinson, G., Tordoff, P., & Drust, B. (2009). Analysis of high intensity activity in

Premier League soccer. Internation Journal of Sports Medicine, 30 (3), 205–212.

Figueroa, P. J., Leite, N. J., & Barros, R. M. L. (2006a). Background recovering in outdoor image sequences: An

example of soccer players segmentation. Image and Vision Computing, 24 (4), 363–374.

Figueroa, P. J., Leite, N. J., & Barros, R. M. L. (2006b). Tracking soccer players aiming their kinematical motion

analysis. Computer Vision and Image Understanding, 101 (2), 122–135.

Frencken, W. G. P., & Lemmink, K. A. P. M. (2009). Team kinematics of small-sided soccer games. In T. Reilly, and

F. Korkusuz (Eds.), Sience and football VI (pp. 161–166). New York: Routledge.

Garganta, J. (1997). Modelacao tactica do jogo de Futebol: Estudo da organizacao da fase ofensiva em equipas de alto

rendimento, Unpublished PhD thesis, Universidade do Porto, Porto.

Golub, G. H., & Van Loan, C. F. (1989). Matrix computations, 2nd ed. Baltimore, MD: Johns Hopkins University

Press.

Football players’ organisation on the pitch 95

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13

Hughes, M. D., & Bartlett, R. M. (2002). The use of performance indicators in performance analysis. [Review]

Journal of Sports Science, 20 (10), 739–754.

McGill, R., Tukey, J. W., & Larsen, W. A. (1978). Variations of box plots. The American Statistician, 32, 12–16.

Misuta, M. S. (2004). Automatic tracking of soccer players trajectories using videogrametry: method validation and results

analysis. Unpublished Master degree thesis, UNICAMP, Campinas.

Misuta, M. S. (2009). Analysis of the automatic tracking of players in collective sports. Unpublished PhD thesis,

University of Campinas, Campinas.

Mitchell, S. A. (1996). Approaches to teaching games: Improving invasion game performance. The Journal of

Physical Education, Recreation and Dance, 2 (2), 30–33.

Okihara, K., Kan, A., Shiokawa, M., Choi, C. S., Deguchi, T., Matsumoto, M., & Higashikawa, Y. (2004).

Compactness as a strategy in a soccer match in relation to a change in offence and defense [Communications to

the Fifth World Congress on Science and Football]. Journal of Sports Sciences, 22 (6), 515.

Preparata, F. P., & Shamos, M. I. (1985). Computational geometry: an introduction. New York: Springer-Verlag.

Yamanaka, K., Hughes, M., & Lott, M. (1993). An analysis of playing patterns in the 1990 World Cup for

Association Football. Science and Football II, 206–214.

Yue, Z., Broich, H., Seifriz, F., & Mester, J. (2008a). Mathematical analysis of a soccer game. Part I: Individual and

collective behaviors. Studies in Applied Mathematics, 121 (3), 223–243.

Yue, Z., Broich, H., Seifriz, F., & Mester, J. (2008b). Mathematical analysis of a soccer game. Part II: Energy,

spectral, and correlation analyses. Studies in Applied Mathematics, 121 (3), 245–261.

F. A. Moura et al.96

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 1

4:26

25

Sept

embe

r 20

13