SIDANE: Towards the Automatic Analysis of

Football Tactics and Actions

D. Vallejo

, G. Alises

, J. A. Albusac

, C. Glez-Morcillo

and J. J. Castro-Schez

School of Computer Science, University of Castilla-La Mancha, Ciudad Real, Spain

School of Mining and Industrial Engineering, University of Castilla-La Mancha, Ciudad Real, Spain

Keywords:

Football, Event Understanding, Expert Systems, Fuzzy Logic, Team Sports.

Abstract:

In recent years technology has been used as an engine to improve the performance of the players and the

preparation offered by trainers in a great variety of sports. In some of them, such as football, technology has

contributed enormously to improving monitoring of players so as to assess their performance and to prepare

tactics and strategies that are so essential nowadays for making the difference, before facing a match. In this

paper we present SIDANE, an expert system capable of detecting important situations in football, from a

tactics point of view and assessing if the players involved took a good decision from the point of view of the

trainer. The SIDANE reasoning engine uses Fuzzy Logic with the aim of tackling uncertainty and facilitating

the deﬁnition of rules close to the way in which the trainer communicates with his or her players. In this

way the distance between expert and machine is greatly narrowed. The results yielded show how powerful

SIDANE is as a starting point for the automatic analysis of important situations in team sports.

1 INTRODUCTION

A few years ago the huge impact that technology has

in the sporting world today seems unthinkable. If,

for example, a famous retired tennis player were to

check ﬁrsthand how Hawk-Eye technology can an-

swer if the ball fell inside or outside the court, he

or she would have wondered, why couldn’t this have

been done before? Indeed, the tentacles of technology

have more than inﬁltrated into any type of sporting

event, whether it be amateur or professional. From

the most basic technology, such as, for example, the

use of digital scoreboards to manage the state of the

game, through to highly complex technological so-

lutions, such as those that create behaviour models

for the players with the aim of improving their per-

formance, technology is an essential support in the

sporting ﬁeld. From a professional point of view, this

support is reﬂected essentially from four perspectives,

which coincide with the main roles adopted in a large

part of the sports there are nowadays: i) player ii)

trainer, iii) referee and iv) spectator.

Professional players use technology to improve

their performance or to train more effectively. For

example a javelin thrower can make use of a move-

ment analysis system to study if the throwing angle

used was the optimum one or not according to other

important parameters such as, for example, his or her

run up speed or wind resistance. A professional bil-

liards player can use an Augmented Reality system

(Azuma et al., 2001) which digitally draws the pos-

sible trajectory of the ball according to the cue direc-

tion. Moreover, all this is done before the ball is hit!

Trainers beneﬁt from automatic tools to ﬁnd out the

players statistics, such as, for example, the distance

covered, or the number of passes successfully made

in ball sports. Furthermore, digital boards are also

habitually used to explain tactics and strategies that

players must follow, thereby saving time with respect

to more traditional alternatives.

From the point of view of the referee, technology

has enormously facilitated the assessment of game

rules thereby avoiding disputes between players and

conﬂicts that arise after seeing a replay on TV of the

problematic move. The aforementioned Hawk-Eye in

tennis is a highly representative example. Photoﬁnish

in athletics or the use of computerized vision tech-

niques to detect if a player is offside (behind the de-

fence line) when playing football (see Figure 1) are

other important examples. Finally, the spectator also

beneﬁts from technology since a highly widespread

trend nowadays consists in superimposing digital in-

formation onto real sports pictures, with occasional

use of Augmented Reality. For example, in sailing

334

Vallejo, D., Alises, G., Albusac, J., Glez-Morcillo, C. and Castro-Schez, J..

SIDANE: Towards the Automatic Analysis of Football Tactics and Actions.

In Proceedings of the 3rd International Congress on Sport Sciences Research and Technology Support (icSPORTS 2015), pages 334-341

ISBN: 978-989-758-159-5

or American football, this approximation is used fre-

quently.

The fruit of our labour presented in this article and

named SIDANE focuses on the role of the football

trainer. Essentially, the aim of SIDANE is to facilitate

the work of these professionals via automatic analysis

of situations in which the players become involved.

However, SIDANE can easily be transferred to other

team sports in which there are a ball to guide how the

game develops.

Nowadays, professional football has developed in

such a way that the concept of team has taken on, if

possible, greater importance. Physical preparation of

elite football players is very demanding and, although

on a technical level a player may shine at speciﬁc mo-

ments, it is the team as a whole that makes the dif-

ference in championships and tournaments of a sig-

niﬁcant duration. In this way, the role of the trainer

takes on special importance when facing matches and

preparing both tactics and strategy for his or her play-

ers. To increase the possibility of success when fac-

ing these tasks, the trainers do not hesitate to use this

new technology with the purpose of communicating

their instructions to the players in the most practical

and efﬁcient way possible. From the technological

support point of view, nowadays there are tools that

are automatically capable of following a player at all

times (D’Orazio and Leo, 2010), measure the distance

covered (Barros et al., 2007), jot down the number of

passes made in a match (Hughes and Franks, 2005),

or even monitor the work proﬁle of the player (Car-

ling et al., 2008).

However, it would be desirable to have higher

level behaviour analysis tools, that is, those that will

study when a player behaves as he or she should when

faced with a certain situation, in keeping with the tac-

tical vision and the strategy of his or her trainer. A

simple example in the football world could be the goal

shot. In this context, a tool that will analyse when a

player should take a goal shot could analyse factors

such as the distance to the goal, the number of op-

ponent players blocking the shot or even, if there is

any unmarked teammate located in a better position

to take a shot at goal. Nowadays, this type of infor-

mation is extracted manually by making notes on the

video recording while the match took place. In this

way, the trainer can justify to his or her players when

they do well and when not. Unfortunately, this task,

normally carried out by the trainers support team, is

very tedious and prone to errors.

This is the main driving force behind SIDANE,

the expert system presented in this work whose aim

is to accurately analyse the actions of football play-

ers. For this purpose, at the core of SIDANE, there

Figure 1: Offside automatic detection.

is a reasoning engine based on rules that determine

how a player should ideally behave in given situa-

tions. The conventionality used in SIDANE is Fuzzy

Logic (Zadeh, 1996), a multivalued logic that allows

approximated reasoning to be carried out, using lin-

guistic labels, instead of exact reasoning. In this way,

rules can be established such as the following: if a

player is surrounded by a certain number of opponent

players, then he or she should pass the ball so as not

to lose it. Fuzzy Logic allows the knowledge neces-

sary for a software system to automatically analyse its

input data, to be shown in a way that is very similar

to how a human expert would do it. This is a highly

important advantage when reducing the distance be-

tween people and machines.

The remainder of this article is structured in the

following manner. In section 2 there is a description

of a study of related work, with special emphasis on

those proposals in which there is an attempt to carry

out some kind of automatic analysis in the world of

football. In section 3 there is a detailed discussion on

what SIDANE consists of as an expert system, what

is its architecture, how its reasoning engine works and

what is the knowledge base that has been used as a

starting point for reﬂecting the instructions given by

the trainer. Then, in section 4 the experimental results

obtained when assessing SIDANE are described. The

article ﬁnishes with section 5, in which the conclu-

sions arrived at are commented on and some future

lines of work are presented.

2 RELATED WORK

The state of the art as regards automatic analysis of

situations in the sporting ﬁeld and, speciﬁcally, in

football, has made plenty of progress in recent years.

In this context, it is worth stressing the importance of

the tracking process as this is fundamental for locat-

ing the players and the ball at every moment (Prince,

2012). This process is carried out by sophisticated

SIDANE: Towards the Automatic Analysis of Football Tactics and Actions

335

Computer Vision systems which analyse the different

pictures or frames that make up the video and which

are captured by a multitude of cameras, which are

strategically placed on the football pitch. In this con-

text, the authors T. D’Orazio y M. Leo made a com-

plete review of the state of the art that the reader may

consult (D’Orazio and Leo, 2010).

From a data obtaining point of view static cameras

may be used or those used for broadcasting matches.

For example, it is common enough to use ﬁxed cam-

eras whose ﬁelds of vision overlap (Xu et al., 2005).

However, it is perfectly possible to use the broad-

casting cameras themselves, such as discussed by M.

Beetz et al (Beetz et al., 2005). These authors con-

sidered an artiﬁcial system used to analyse the 2006

World Cup matches with the purpose of positioning

the players at all times. On this point, it must be em-

phasized that regardless of the data source, there are

traditional problems that any algorithm must tackle,

such as, for example the ever present occlusions or

variations in the lighting sources. One work which

is representative of an attempt to tackle this problem

is proposed by S.H. Khatoonabadi and M. Rahmati

(Khatoonabadi and Rahmati, 2009), where they con-

sidered an algorithm made up of different steps, from

a lower to higher level of abstraction, so as to, ﬁnally,

obtain the position of every player on the game ﬁeld.

Another large branch of activities that revolves

around the tracking process is concerned with the

use of hardware devices as a common denominator,

which are incorporated into the player itself or the

ball, instead of analysing pictures. For example, it

is possible to use miniscule transmitters to directly,

and with little scope for error, obtain the position of

the players (Rohmer et al., 2001). Essentially, this

approach is closely linked to GPS systems.

From an architecture perspective this layer of

tracking is used as a base for other layers that have a

greater level of abstraction whose aim is a higher level

analysis. Then, for example, the position of the play-

ers and that of the ball could be used to feed an intelli-

gent algorithm which will automatically determine in

which position of the game ﬁeld the player is making

most effort. Another more complex example could be

based on an algorithm that automatically obtains the

degree of pressure from the team that does not possess

the ball at each moment. Essentially, the trainer of a

football team can use this knowledge to improve the

preparation of tactics and strategy for his or her team,

thereby saving a great deal of time as regards the tra-

ditional alternative which consists of viewing videos

and making commentaries manually.

A signiﬁcant amount of researchers are interested

in obtaining statistical data that allows the tactical in-

formation to be used by the trainers to be extrapolated

so as to improve the performance of their team. For

example, G. Zhu et al considered the concept of added

trajectory to extract tactical information in situations

in which a team has scored a goal (Zhu et al., 2007).

Another large ﬁeld of study revolves around the anal-

ysis of the individual skills of the players. In this con-

text, an important item is possession of the ball, since,

a priori, greater possession of the ball implies greater

control of the game and, therefore, greater probabil-

ity of winning. To minimize the manual work de-

rived from this task, X. Yu et al put forward a semi-

automatic system capable of calculating ball posses-

sion by each player from the information obtained

from the television cameras (Yu et al., 2005). Other

related work is discussed by P.S. Tsai et al, which

is focused in this case on analysing the speed of the

players during the game (Tsai et al., 2007).

Beyond this individual study, the system put for-

ward by M. Beetz et al is worth mentioning, in which

the real time video streaming analysis is considered in

order to recognize activities and events (Beetz et al.,

2005). This concept of event is especially important

for the state of the art of this type of systems to make

progress. With the work mentioned above, the system

designed uses a reasoning engine based on ﬁrst order

logic to determine if a situation that was learnt pre-

viously has taken place in the match under analysis.

For example, the authors comment on the examples

of events the opportunity of scoring a goal or the situ-

ation in which a player is under pressure. The system

maintains a series of rules that are ﬁred to calculate

the probability of each event handled by the situation

occurring in relation to the real situation under analy-

sis.

Other works related with the above approach are

discussed in the literature (Miene et al., 2004), (Nair

et al., 2004). Both work in this case with sim-

ulated data obtained from the famous competition

Robocup

, in which there are leagues both for physi-

cal robots and for virtual players which are simulated

by means of sophisticated algorithms. It is worth em-

phasizing that in these works events such as the aver-

age distance covered or possession by each player to

automatically analyse if a situation is to his or her ad-

vantage or not, are studied and detected. Another rep-

resentative event, associated with automatical assess-

ment of the offside event, is discussed by T. D’Orazio

et al (D’Orazio et al., 2009). This work is orientated,

however, to facilitating assistance to referees when as-

sessing if there is an offside situation or not.

http://www.robocup.org

PerSoccer 2015 - Special Session/Symposium on Performance Analysis in Soccer: How does Technology Challenge Current Practices? -

3rd Edition

336

GUI

Data processing

module

Action

assessment

sub-module

Situation

detection

sub-module

Tracking

system

Automatic analysis module

Knowledge

Base

Human

Expert

Fuzzy

Logic

Figure 2: Architecture overview of SIDANE.

3 AUTOMATIC ANALYSIS OF

EVENTS

3.1 Overview

SIDANE is an expert system for the analysis and au-

tomatic assessment of situations and events in the

world of football. As will be discussed further on,

SIDANE is capable of, for example, detecting when

a player must pass the ball and when not. For this

purpose, SIDANE makes use of an expert knowledge

base, deﬁned taking into account the instructions from

the football trainer, and an inference system based

on rules that determine when a player is behaving

well and when not. Figure 2 graphically shows the

SIDANE architecture taking into account the stream

of information, that is, from obtaining raw data, using

for, example cameras, until high level knowledge is

created such as that commented upon in the previous

example.

Initially, the tracking system is that responsible for

obtaining the position of all the players and of the

ball in the games area at every moment. In other

words, the responsibility this system has could con-

sist in transforming the video streams captured by the

cameras into 3D information which shows the posi-

tion of the players and the ball. The tracking system

is outside the scope of this paper, but the SIDANE

architecture has been designed to guarantee incorpo-

ration of any tracking system. Below, the data pro-

cessing module normalizes the information relative to

said positions, in such a way that they are all found

to be in the range [0.0, 0.0, 0.0] and [1.0, 1.0, 1.0].

Moreover, this sub module stores all the information

received by the tracking system in a database with the

purpose of making, if necessary, a forensic analysis of

the data stored.

The automatic analysis module is the real heart

of SIDANE and is made up of a reasoning system

capable of detecting and analysing interesting foot-

ball situations from a tactical and strategical point of

view. For this purpose, and as shown in ﬁgure 2, this

module uses a previously deﬁned knowledge base that

houses a series of rules. These rules determine if the

ideal behaviour of the football player is in accordance

with the criteria of his or her trainer. The mathe-

matical conventionality that sustains both the reason-

ing system and the knowledge base is Fuzzy Logic

(Zadeh, 1996). Said conventionality is an extension

of Boolean logic that allows dealing with uncertainty

and vagueness of real world problems. For example,

from a semantics point of view it is more practical to

state that the ball is close to the goal than state that the

ball is 1.34 metres from it. In the following section

this conventionality will be studied in greater depth.

This behaviour analysis module encompasses, in

turn, two other sub modules. The ﬁrst of these is the

situations detection sub-module, responsible for de-

tecting what is happening in the game area. For ex-

ample, the reader may think of the classical offside

situation. The second sub-module is the actions as-

sessment sub-module, responsible for assessing if the

player took the right decision in accordance with that

deﬁned by the knowledge base and within the context

of the previously detected situation. For example, the

reader may consider the situation in which the central

defence remains back, so offside does not apply and

the forward of the opponent team is enabled to act.

SIDANE: Towards the Automatic Analysis of Football Tactics and Actions

337

Finally, SIDANE provides a user graphics inter-

face that enormously facilitates the detection of er-

rors made by the players and that, additionally, pro-

vides detailed information of their states at every mo-

ment. It is worth emphasizing that the SIDANE archi-

tecture has been designed with special consideration

for scalability maintaining a high degree of indepen-

dence among the modules inserted and deﬁning some

simple interfaces that enable other modules to be in-

corporated.

3.2 Knowledge Base and Reasoning

Engine

Fuzzy Logic (Zadeh, 1996), put forward by professor

Zadeh in 1965, has been the conventionality chosen as

the focal point for SIDANE. Essentially, the main fea-

ture of Fuzzy Logic is it allows the quantiﬁcation of

imprecise values of our language, such as much, little,

or too much, adapting itself better to real world prob-

lems than traditional logic, which only allows for two

possible values: true (1) and false (0). In this context,

the human brain has a great ability to interpret and

solve complex situations without needing to handle

numerical values. It is precisely from this consider-

ation, that the computing with words stream arose, to

which Fuzzy Logic adapts perfectly with the concepts

of linguistic variables and fuzzy rules.

In the problem dealt with in this paper, a linguis-

tic variable could be the attitude of the player which,

in turn, could take the values of defensive, neutral,

offensive and very offensive, for example. Subse-

quently, this variable could be used in a fuzzy rule

of the type ”IF the attitude of the player is very of-

fensive AND the distance to the goal is short, THEN,

make a goal shot”. This knowledge deﬁnition model

is signiﬁcantly close to the expert in the domain, that

is, the trainer, which drastically reduces the gap be-

tween the latter and the machine. In this paper the

choice of Fuzzy Logic as a knowledge representation

conventionality is complemented with the inference

method of Mamdani (Mamdani, 1974). Essentially,

this method is responsible for i) changing numerical

values to fuzzy values, ii)making the reasoning pro-

cess fuzzy and ﬁnally, iii) changing the fuzzy values

obtained into numerical values that can be interpreted

once again by a machine.

Figure 3 shows on the left hand side the fuzzy di-

vision of the game area into different areas according

to two independent linguistic variables: X Pos and

Y Pos. In this way, a forward that attacks the goal

on the right will normally be situated in the values

Right or Very Right of the Y Pos variable, with less

likelihood of retreating to defend and that, therefore,

the Y Pos variable takes values of Medium, Left or

Very Left. This type of basic variables are the ﬁrst

level of the knowledge base deﬁned in this paper. The

deﬁnition of another example of a basic variable de-

nominated Attitude, is shown in the upper right part

of ﬁgure 3. Here, the possible values of said variable

are Defensive, Neutral and Offensive.

The rules of the knowledge base of this ﬁrst level

are used to determine the occurrence of states or basic

actions. Rules 3 and 4 of ﬁgure 3 show two basic ex-

amples. With rule 4, SIDANE is capable of inferring

that if a player has possession of the ball and the num-

ber of opponents that are close to him or her is high,

then the pressure level is high. The SIDANE knowl-

edge base maintains more rules to determine the dif-

ferent pressure values according to if the number of

close opponents is high or not. See how these type

of rules are really simple to express for the trainer,

which hugely facilitates their incorporation into the

SIDANE fuzzy knowledge base.

Now pay attention to rule 10 shown equally in ﬁg-

ure 3. This rule may seem more complex because the

number of antecedents is greater, but in reality it is

also rather simple. Rule 10, together with other simi-

lar rules included in the knowledge base, are used so

that SIDANE infers when a player is found in an area

of the ﬁeld where it is suitable to make the so called

killer pass and when not. This classic footballer move

consists in passing the ball to a forward in such a way

that the latter can score a goal just by pushing the ball.

Obviously, this is a highly desirable situation. The

rule in question assesses whether the one passing is

lightly hugging a sideline and, moreover, is close to

the rival goal. If this is the case, then the player is

located in an area in which said pass is a desirable

option.

The SIDANE knowledge base handles a second

level or group of rules in which football concepts fre-

quently worked on in football team training sessions

are managed, such as for example dangerous loss of

ball, goal shot or long pass. SIDANE is capable

of automatically detecting when these situations take

place and assessing if they were made at the right time

or not. The actions associated with these concepts

usually show yes or no type decisions, so deﬁning

them is simple. Figure 3 shows on the right hand

side the deﬁnition of two variables associated with

these actions: killer pass and goal shot. However,

the most interesting part is in rule 16. This second

level rule uses as an antecedent the information ob-

tained by the ﬁrst level rules (e.g. player situated in

the killer pass area or attitude of the player) to deter-

mine if the player must carry out the killer pass or not.

The example shows if the player has the ball, is in the

PerSoccer 2015 - Special Session/Symposium on Performance Analysis in Soccer: How does Technology Challenge Current Practices? -

3rd Edition

338

Very Left Very RightLeft

Medium Right

Very Down Down

Medium Up Very Up

Examples of fuzzy variables

DEFUZZIFY Attitude

TERM Defensive := (0, 0) (0, 1) (0.25, 1) (0.50, 0);

TERM Neutral := (0.25, 0) (0.50, 1) (0.75, 0);

TERM Offensive := (0.50, 0) (0.75, 1) (1, 1) (1, 0);

METHOD : COG;

DEFAULT := 0;

END_DEFUZZIFY

DEFUZZIFY KillerPass

TERM Yes := 1;

TERM No := 0;

METHOD: COGS;

DEFAULT := 0;

END_DEFUZZIFY

Examples

of fuzzy

rules

RULE 3: IF (X_Pos IS Right) OR (X_Pos IS VeryRight)

THEN Attitude IS Offensive;

RULE 4: IF (Possession IS Yes) AND (Nearly_players IS High)

THEN Pressure IS High;

RULE 10: IF ((X_Pos IS VeryLeft OR X_Pos IS Left) OR (X_Pos IS VeryRight OR X_Pos IS Right)) AND (Y_Pos IS Down OR Y_Pos IS Up)

THEN KillerPassZone IS Yes;

RULE 16: IF (Possession IS Yes) AND (KillerPassZone IS Yes) AND (Attitude IS Offensive) THEN KillerPass IS Yes;

DEFUZZIFY Shoot

TERM Yes := 1;

TERM No := 0;

METHOD: COGS;

DEFAULT := 0;

END_DEFUZZIFY

Figure 3: Left: Fuzzy division of the soccer ﬁeld. Right: Examples of fuzzy variables. Below: Examples of fuzzy rules in

the knowledge base. The adopted notation is related to the Fuzzy Control Language (FCL).

suitable position and his or her attitude is offensive,

then he or she should make a pass that will potentially

leave a teammate alone so the latter can score a goal.

It is precisely in this type of rule where the essence

of the expert system put forward in this work can be

appreciated- SIDANE is capable of assessing if the

decision taken was the correct one or not, considering

the action of the player and the trainer criteria which

is reﬂected in the SIDANE knowledge base.

4 RESULTS

Our ﬁrst aim for assessing the behaviour of SIDANE

consisted in obtaining real football match data. Un-

fortunately, there is no public repository with the

players tracking data and the research groups whom

we contacted were not able to share theirs due to pri-

vacy laws. For this reason, in order to assess SIDANE

we have used data from the RoboCup 2D Soccer Sim-

ulation League

. This league is a worldwide compe-

tition in which a range of teams of independent soft-

ware agents compete to be crowned as the best. The

software used in this competition allows tracking data

to be obtained for all players and from the ball, so it is

perfect to assess SIDANE. Likewise, the level of so-

phistication of the virtual players is so high that they

http://en.wikipedia.org/wiki/RoboCup 2D Soccer Sim-

ulation League

behave as if they were real players. In this paper data

has been used from the ﬁnal of the RoboCup 2012

between the Helios (Japan) and WrightEagle (China)

teams.

Figure 5 shows the number of detected and auto-

matically assessed SIDANE situations. Speciﬁcally,

these situations are the following: area pass, danger-

ous loss of ball, killer pass and, ﬁnally, goal shot. On

this point it must be clariﬁed that the dangerous loss

of ball situation is that which triggers a counterattack

by the rival team. In other words, it is a situation to be

avoided by the team that possesses the ball. The long

pass refers to a situation in which, for example, it is

desirable to move the ball to the sideline opposite the

game area in which, potentially, there are fewer op-

ponent players and, therefore, more opportunities to

advance.

Moreover, ﬁgure 4 shows two situations shown by

the SIDANE graphic interface which allow what has

been detected and what the players should have done

to be viewed rapidly. Figure 4.a shows the red player

in possession of the ball and a considerable number

of opponent blue players surrounding him or her. In

other words, SIDANE shot the rule in which there is

a possibility of dangerous loss of ball by the other

blue team. Here, the red player should pass the ball

or throw it out of range of the other team to not lose it

when faced with such a dangerous position. SIDANE

detected and assessed 30 dangerous loss of ball sit-

SIDANE: Towards the Automatic Analysis of Football Tactics and Actions

339

(b)(a)

Figure 4: a) Dangerous loss of ball detection. b) Desirable long pass situation.

uations in the test match. Another especially impor-

tant situation from the trainer point of view is shown

in ﬁgure 4.b. Here, it is desirable that the red player

that has the ball makes a large run with the purpose

of breathing some fresh air into the game and begin-

ning the attack against the opponent goal. SIDANE

is also capable of detecting and assessing these types

of situations. In the test match, SIDANE detected and

assessed 90 situations in which a long run was desir-

able.

Before discussing the conclusions and the future

potential lines of work associated with SIDANE the

reader is advised to see a video in which situations

obtained from real matches are contrasted with sit-

uations analysed by SIDANE in the RoboCup 2D

Soccer Simulation League. This video is available

at http://www.esi.uclm.es/www/dvallejo/sidane/video.

Figure 5: Detected and assessed situations by SIDANE in

the RoboCup 2012 ﬁnal.

When SIDANE detects and assesses a situation re-

ﬂected in its knowledge base, the graphics interface

marks the player involved with a blue circle, if the

right decision is taken, or with a red circle if the wrong

one is taken in accordance with that previously de-

ﬁned by the trainer in the knowledge base.

5 CONCLUSIONS AND FUTURE

WORK

SIDANE is an expert system capable of automatically

detecting and assessing essential situations from the

point of view of the football trainer. Essentially, the

main contribution this paper has made is in the use

of expert knowledge to detect and assess high level

situations and events in the football world. This con-

sideration is a step forward with respect to the tradi-

tional techniques of note taking which is usually done

manually and revolves around the trainers assistants

watching videos to select the most important moves.

To reach its objective, SIDANE has at its core a rea-

soning system based on Fuzzy Logic which makes

use of a knowledge base made up of variables and

rules that reﬂect the tactical and strategical criteria of

a human expert or trainer. The use of If-Then rules

greatly facilitate the modelling of said criteria and

simplify their incorporation into the SIDANE knowl-

edge base. Moreover, the SIDANE architecture has

been designed to incorporate any tracking system and

to extend its knowledge base with new rules associ-

ated with a great number of concepts and footballing

situations.

PerSoccer 2015 - Special Session/Symposium on Performance Analysis in Soccer: How does Technology Challenge Current Practices? -

3rd Edition

340

To assess SIDANE we have used data obtained

from the RoboCup 2D Soccer Simulation League,

since it was possible to directly know the results of the

tracking process of the players and of the ball at ev-

ery moment. The analysis of a full match has allowed

the identiﬁcation of a large number of situations that

revolve around concepts that are so essential from a

tactical perspective such as the area pass, dangerous

loss of ball, killer pass, long pass or goal shot.

Currently, our work is focused on detecting and

assessing collaborative events in which various play-

ers are involved in the same situation. Speciﬁcally,

one of the situations in which we are particularly in-

terested is the offside trap, used by the defence of a

team to leave the forward of the opponent team off-

side and, in this way, retake possession of the ball

directly. Furthermore, we continue to search for data

from real matches and hope that this paper contributes

to us being able to carry out this task successfully.

ACKNOWLEDGEMENTS

The authors would like to thank the University of

Castilla-La Mancha, for funding this work under the

research project GI20153014.

REFERENCES

Azuma, R., Baillot, Y., Behringer, R., Feiner, S., Julier, S.,

and MacIntyre, B. (2001). Recent advances in aug-

mented reality. IEEE Computer Graphics and Appli-

cations, 21(6):34–47.

Barros, R., Misuta, M., Menezes, R., Figueroa, P., Moura,

F., Cunha, S., Anido, R., and Leite, N. (2007). Anal-

ysis of the distances covered by ﬁrst division brazil-

ian soccer players obtained with an automatic track-

ing method. Journal of Sports Science & Medicine,

6(2):233.

Beetz, M., Kirchlechner, B., and Lames, M. (2005). Com-

puterized real-time analysis of football games. IEEE

Pervasive Computing, 4(3):33–39.

Carling, C., Bloomﬁeld, J., Nelsen, L., and Reilly, T.

(2008). The role of motion analysis in elite soccer.

Sports Medicine, 38(10):839–862.

D’Orazio, T. and Leo, M. (2010). A review of vision-based

systems for soccer video analysis. Pattern Recogni-

tion, 43(8):2911–2926.

D’Orazio, T., Leo, M., Spagnolo, P., Mazzeo, P., Mosca, N.,

Nitti, M., and Distante, A. (2009). An investigation

into the feasibility of real-time soccer offside detec-

tion from a multiple camera system. IEEE Transac-

tions on Circuits and Systems for Video Technology,

19(12):1804–1818.

Hughes, M. and Franks, I. (2005). Analysis of passing se-

quences, shots and goals in soccer. Journal of Sports

Sciences, 23(5):509–514.

Khatoonabadi, S. and Rahmati, M. (2009). Automatic soc-

cer players tracking in goal scenes by camera motion

elimination. Image and Vision Computing, 27(4):469–

479.

Mamdani, E. (1974). Application of fuzzy algorithms for

control of simple dynamic plant. In Proceedings of

the Institution of Electrical Engineers, volume 121,

pages 1585–1588.

Miene, A., Visser, U., and Herzog, O. (2004). Recognition

and prediction of motion situations based on a quali-

tative motion description. In RoboCup 2003: Robot

Soccer World Cup VII, pages 77–88. Springer.

Nair, R., Tambe, M., Marsella, S., and Raines, T.

(2004). Automated assistants for analyzing team be-

haviors. Autonomous Agents and Multi-Agent Sys-

tems, 8(1):69–111.

Prince, S. (2012). Computer Vision: Models, Learning and

Inference. Cambridge University Press.

Rohmer, G., D

unkler, R., K

ohler, S., and von der Gr

un, T.

(2001). A microwave based tracking system for soc-

cer. In Proceedings of the 18th International Techni-

cal Meeting of the Satellite Division of The Institute of

Navigation (ION GNSS), pages 2207–2212.

Tsai, P., Meijome, T., and Austin, P. (2007). Scout: a game

speed analysis and tracking system. Machine Vision

and Applications, 18(5):289–299.

Xu, M., Orwell, J., Lowey, L., and Thirde, D. (2005). Ar-

chitecture and algorithms for tracking football players

with multiple cameras. IEE Proceedings-Vision, Im-

age and Signal Processing, 152(2):232–241.

Yu, X., Hay, T., Yan, X., and Chang, E. (2005). A player-

possession acquisition system for broadcast soccer

video. In IEEE International Conference on Multi-

media and Expo (ICME), pages 522–525.

Zadeh, L. (1996). Fuzzy logic = computing with words.

IEEE Transactions on Fuzzy Systems, 4(2):103–111.

Zhu, G., Huang, Q., Xu, C., Rui, Y., Jiang, S., Gao, W., and

Yao, H. (2007). Trajectory based event tactics analysis

in broadcast sports video. In Proceedings of the 15th

International Conference on Multimedia, pages 58–

67.

SIDANE: Towards the Automatic Analysis of Football Tactics and Actions

341