
Abstract—One of the most heavily debated issues in the world
of sport is whether technology can positively impact a sport
without being intrusive. In the world of soccer, the issue of
incorrectly awarding or not awarding a goal has been a source of
significant controversy. To reduce the chance of making a wrong
call, goal-line technology (GLT) systems have been proposed and
implemented at the highest level of the game. This report
introduces the concept of GLT, discusses the image processing
techniques that play a big part in a successful GLT system,
namely object detection and object tracking, and explains the
working of two GLT systems used in soccer that use image
capture technology to make split-second decisions, namely the
systems proposed by Hawk-Eye and GoalControl.
Index Terms—Soccer, goal-line technology, ball tracking,
object detection, object tracking, Hawk-Eye, GoalControl
I. INTRODUCTION
OCCER is the world’s most popular sport. Viewership for
the 2014 FIFA World Cup Final in Brazil was estimated
at 1.01 billion, the single largest for any sporting event ever
recorded. Such a large global presence and following
automatically places intense scrutiny upon the most fleeting
of incidents on the soccer field. Thousands of crazy fans jump
up in protest or gleeful cheering when the referee makes a
controversial decision that might impact the game in a huge
way, and there are several examples of such incidents
throughout the course of soccer history. Due to the speed at
which the modern game is played, the possible presence of
obstacles that might hinder the referees’ views, the intense
pressure on referees, human error and many other factors, the
frequent occurrence of such incidents during the course of a
90-minute soccer game is not altogether surprising.
S
With the advent of modern technology, an increasing
number of voices have called for the introduction to soccer of
cutting-edge devices and gadgets equipped with tools to make
these controversial decisions on behalf of the referee.
Arguably the most important of these decisions involve
whether a goal has been scored. In order reduce human
involvement in the matter of goals, several proposals have
been made to install goal decision systems, ranging from a
chip on the ball to a magnetic field within the area enclosed
by the goalposts.
Two proposals – Hawk-Eye and GoalControl – have
emerged as front-runners in the business. Both these systems
employ a number of high-speed cameras, installed at various
positions around the stadium and pointing towards one of the
two goals, that detect whether the ball has crossed the goal
line or not. This report includes a significant understanding
of these systems, and their approaches to determine the
legitimacy of a goal.

II. BACKGROUND AND MOTIVATION
An inherent quality of sport is that it is unpredictable; that
is the beauty of sport. The human element involved in sport
inherently creates several situations where a judgment call
must be made by the official. A tennis ball that lands out
might be called “in”; a baseball batter might strike out despite
a clear “ball” pitch; a basketball team might be awarded
possession despite themselves having touched the ball last
before it went out of bounds. While judgment calls such as
these are more often than not a part of the game, even the
smallest such incident might have severe repercussions due to
the competitive nature of sport.
The famous 1966 FIFA World Cup final between England
and Germany was controversially decided by one of the most
famous judgment calls in all of soccer. As the game reached
its final stages, 11 out of 30 extra-time minutes had been
played with the score tied at 2-2 when England striker Geoff
Hurst struck the ball against the vertical bar; the ball landed
square on the line (see Fig. 1) before being cleared by a
defender.
Fig. 1: Geoff Hurst's goal during the 1966 FIFA World Cup final
Due to the lack of available technology and in accordance
with the official rules, the referee and the linesman convened
to discuss their individual points of view, and subsequently
awarded a goal to the England team; despite angry reactions
from the German players, the goal stood. England went on to
win their first World Cup. Retrospective analysis shows that
the ball did not fully cross the line, meaning the goal should
not have been allowed.
Can such a significant error be corrected today? With
significant research and improvements in the field of image
processing and vision, as well as the availability of super
high-speed cameras to track even infinitesimal incidents, the
answer is a resounding yes.
1
Goal-Line Technology in Soccer:
Discussion and Evaluation of Systems
Arvindh Mani, UFID: 57398904
III. LITERATURE REVIEW
A. Goal-line technology
In simple terms, goal-line technology (GLT) in soccer is a
way to determine whether a goal has been scored. A goal is
scored when the ball is anywhere within the confines of the
two vertical poles, known as goalposts, and one horizontal
pole, known as the crossbar.
As per FIFA’s GLT testing manual, “The objective of goalline
technology is not to replace the role of the officials, but
rather to support them in their decision-making. The GLT
must provide a clear indication as to whether the ball has
fully crossed the line, and this information will serve to assist
the referee in making his final decision.” The guidebook also
states the four basic requirements a successful GLT system
must fulfill:
 The goal-line technology applies solely to the goal line
and only to determine whether a goal has been scored
or not;
 The GLT system must be accurate;
 The indication of whether a goal has been scored must
be immediate and automatically confirmed within one
second; and
 The indication of whether a goal has been scored will
be communicated only to the match officials (via the
referee’s watch, by vibration and visual signal).
B. Use of technology in other sports
Cricket and tennis are two sports that have embraced
technology in order to assist officials in making judgment
calls. Cricket has employed TV replays since the early 1990s
to accurately determine run-outs, and more recently has
included Hawk-Eye technology in its repertoire to determine
the outcome of an LBW decision. Tennis has used Hawk-Eye
since 2002 to accurately determine the landing position of the
ball and reduce human error.
The stop-start nature of both cricket and tennis is a great
boon to technological intervention. Since both sports have
seconds-long “plays” separated by pauses, a decision review
system can be called for between plays without severely
affecting the natural flow of the game. Sports such as
basketball, baseball and football are also conducive to
technological intervention in between plays for the same
reason. Unfortunately, due to the non-stop nature of soccer,
the constraints placed on automated systems are far more
strict. One of these constraints pertains to the time taken to
reach a conclusive decision. Since plays are continuous
without delay as well as uninterrupted, the GLT system
employed in soccer must not only be able to instantaneously
reach a conclusion, but do so in a non-intrusive manner.
C. External factors affecting GLT systems
The GLT system must be able to eliminate external factors
such as player obstruction, partial visibility of the ball, etc.
The presence of several players on the goal line is a common
one, and at no point must the GLT be unable to make a
decision. This can be resolved by, for instance, having
multiple cameras working simultaneously so that a
combination of images obtained by these cameras can be used
to accurately predict the exact location of the ball at a given
point in time (Hawk-Eye, GoalControl). Other approaches
include using sensor-based technology to determine the
location of the ball (GoalRef), installing sensors inside the
ball (Cairos), and more.
IV. OVERVIEW OF A GOAL-LINE TECHNOLOGY SYSTEM
Fig. 2a shows a schematic diagram of a general goal-line
technology system. Four high-speed cameras are placed on
the four goal sides in such a way that their optical axes are
parallel to the surface plane. These cameras are capable of
capturing images at 200 fps. Each camera has a processor
connected to it. The processor is responsible for recording and
processing the images being taken by the camera. A main
node, which has the supervisor function, is connected to all
four camera processors. It is responsible for evaluating and
comparing the results obtained from the four processors. It
also sends the alarm signal to the referee’s watch when a goal
has been detected.
A. Processing nodes
Fig. 2b shows a schematic diagram of the processing steps
at a camera processing node.
These steps are explained below:
2
Fig. 2a: Schematic diagram of a GLT system
Fig. 2b: Flow diagram of the processing steps at a
node in the GLT system
i. Initial calibration: A set of four points, one for each
camera on the field, is taken so as to observe the
correspondence between the ground plane and the
image plane. A transformation matrix M is evaluated
that relates the set of points on the image plane with
those on the real plane. The point where the two
viewing lines intersect is the point estimate of the ball
(see Fig. 3).
From the formula of transformation of projections from
one plane to another, we have
(1)
or, x = MX. In Cartesian coordinate form, this
transformation becomes
(2)
where (u,v) are the coordinates in the image plane and
(X,Y) are the coordinates in the real plane. The
transformation matrix M is obtained by solving
equation (1) using the correspondence between the
(u,v) plane and the (X,Y) plane of the four points.
Repeating this for the two cameras on the two image
planes, the projections (P1,P2) on the real plane can be
obtained (Fig. 3). The intersection of (C1,P1) and
(C2,P2) is the real position of the ball on the field.
Then, it is only necessary to calibrate the positions of
the two cameras (C1 and C2) and the positions of the
four reference points once, when the cameras are
installed. If the cameras are fixed, these calculations
are valid at all times.
ii. Moving object segmentation: This process is very
essential to a GLT system, as it focuses the area where
the ball is to be searched, and also reduces the time for
computation. A segmentation algorithm analyzes the
connected regions from the input image set, allowing
for the evaluation of the size of regions and hence the
possibility of eliminating large sample spaces where
the ball search need not be applied. A background
subtraction algorithm is used to distinguish moving
points from static points.
iii. Circle detection: The circle detection process is the
central process in building a GLT system. The
algorithm to identify the soccer ball has to be fast,
simple and effective. To do this, the proposed method is
split into two parts: a) a general circle detection step;
and b) a ball recognition step. The circle detection
algorithm being used is based on Circle Hough
Transform (CHT) (Fig 4). The CHT aims to find
circular objects of a given radius from the input image
set. A number of modifications have been implemented
on the circle detection algorithms in the last few years,
in order to reduce the burden on computation and
reduce the number of false positives.
iv. Ball recognition: Once circles have been detected, they
are passed on to a neural classifier that has been
trained to separate objects as “ball” and “no ball”
objects. Fig. 5 contains examples of objects accepted by
the ball recognition process.
v. Ball tracking: This procedure is applied to the regions
that were earlier estimated to contain the ball. Once the
ball detection algorithm identifies the ball, a tracking
procedure is initiated. During this phase, the ball
search is applied to all moving regions of the image,
with no regard for their position. Then, in each
successive image, the ball search algorithm is
constrained by a small area around the ball to speed up
this step. If the ball is not found in the predicted ball
position, the search is repeatedly performed on an
increasingly larger area until the ball is found, and the
tracking step continues.
B. Supervisor node
The supervisor node is responsible for making decisions
regarding ball detection and tracking, in accordance with the
data obtained from each camera. The supervisor node
analyzes the data obtained to decide between the following
events: Goal, No Goal, Probable Goal and Line Passing.
3
Fig. 3: The intersection of the lines formed by the four reference
points produces the point estimate of the ball
Fig. 5: Different appearances of the ball
Fig. 4: Hough Transform algorithm for circles
Another aspect that the supervisor node must take into
account is the position of the ball with respect to the goal
plane. For a goal event to be detected, the ball must be
detected to be beyond the goal line and inside the goal posts.
However, due to the fact that homographic transformations
might be affected by small errors in the areas near the
woodwork, an uncertainty region is also defined, where the
system cannot be sure about the position of the ball. This
region is called the “probable goal” region.
Table 1 gives the summary of decisions taken by the
supervisor node.
C. Summary of goal-line technology system
The above is a simple model of a GLT system that, in
practice, can detect the event of a goal or no goal with
significant accuracy. By developing more robust algorithms,
faster and more powerful hardware and reducing the chance
of errors by increasing the number of hardware components, a
very high accuracy can be achieved, making it a foolproof
indicator of goal decisions.
There are a few drawbacks, the main one being the cost of
installation and operation. The high-speed cameras need to
capture data at a rate of 200 frames per second. Over a 90minute
game, that adds up to 1.08 million images. These
images also need to be processed at the supervisor node. The
cost of each camera is estimated to be around ₤1,500. In total,
it is estimated that the cost incurred at each stadium would be
a minimum of ₤250,000 for initial setup and additional costs
for running the system, depending on which system is
implemented. Such astronomical figures have been major
hindrances to professional soccer leagues around the world,
most notably in Germany, where the league refused use of the
available technology citing cost as the prime factor.
V. HAWK-EYE AND GOALCONTROL
We now come to two GLT systems that have received
acceptance by FIFA to be installed across stadiums. Both
these systems use a more enhanced version of the above GLT
prototype, using seven high-frame rate cameras permanently
pointed towards each goal, bringing the total to 14 per
stadium. Both systems approach the problem of goal detection
in a similar way. Minor differences lie in the positioning of
cameras, and the function of the central unit.
A. Hawk-Eye
Hawk-Eye’s foray into the sports video analysis domain
began in 2001, when it was developed in the United Kingdom
for television purposes in cricket. Originally intended to be
used to track the trajectory of the cricket ball in flight, it was
later adopted to be used as a support system for deferring
contentious decisions to the “third umpire”, an official sitting
in a video control room and equipped with systems to assist
him in making the correct decision. One of the most useful
applications early on was in judging leg-before-wicket
decisions correctly.
Hawk-Eye’s impact in the world of professional tennis has
been massive. Starting from 2004, Hawk-Eye’s AutoRef
system has been used to identify the exact landing position of
the ball. Today, it has been used in almost every level of
tennis, and has become a ubiquitous part of the game, with
the challenge rule that facilitates Hawk-Eye’s AutoRef system
being incorporated officially into the ATP tour.
Hawk-Eye’s introduction to the world of soccer was in 2012,
when it was approved (along with GoalRef) and advanced to
a second phase of testing. The technology was first tested in
an international friendly between Belgium and England at
Wembley, although the data was not available for use during
the game; only FIFA had access to the system readings. In
April 2013, technology was accepted at the highest levels of
professional soccer when Hawk-Eye was officially approved
by the Premier League for use in its 2013-14 season. The first
use of Hawk-Eye was on 17 August 2013 at Anfield,
Liverpool. The first time Hawk-Eye verified a goal was on 18
January 2014 at the Etihad Stadium, Manchester.
B. GoalControl
GoalControl is a technology company based in the district
of Aachen, Germany. GoalControl’s technology was licensed
for the first time by FIFA in 2013. Upon successful testing of
the camera systems employed in this GLT system, it was used
for the first time in the 2013 FIFA Confederations Cup in
Brazil where it was used to detect 68 goals throughout the
competition. Later, during the 2013 Club World Cup held in
Morocco, GoalControl was used only to track the ball
(GoalRef was used for display).
In early 2014, a vote was cast by 36 members of the top two
divisions of the German Bundesliga. Needing 24 for the
motion to carry, only 12 of the 36 members voted in favor of
including goal-line technology, citing the exorbitant cost of
installation and maintenance per club.
In 2014, history was made when GoalControl became the first
company to license a GLT system at the World Cup, when it
was officially featured as the review system to be used in the
2014 FIFA World Cup in Brazil. On 15 June 2014, in a game
between France and Honduras at Porto Alegre, a shot from
France’s Karim Benzema bounced back off the goalpost, and
an ensuing fumble from the goalkeeper led to an own goal.
This incident was the first time GoalControl’s system was
used to award a goal.
C. Working of the systems
Both Hawk-Eye and GoalControl employ seven high-speed
cameras, capable of capturing images at a rate of nearly 500
4
System Decision
Node 1
Answer
Node 2
Answer
Homo-
graphic
Control
Position
Control
GOAL Ball Ball Ball Inside
NO GOAL Ball Ball Ball Outside
PROBABLE GOAL Ball Ball Ball
Probably
Inside
LINE PASSING
Ball Ball No Ball —
{Ball, No
Ball}
{No Ball,
Ball}
— —
Table 1: Summary of supervisor node decisions
frames per second. The cameras are constantly detecting
movement on the field, and must, with a very high level of
accuracy, be able to distinguish between the ball and a human
object.
At the moment the ball enters the penalty area, designated as
the bigger rectangular box enclosing the goal space, the
cameras become active and transmit images of the ball to the
central unit. This unit is responsible for making the final
decision when the ball is in the vicinity of the goal. If the ball
crosses the goal line, the central unit sends an encrypted
signal to the match official’s watch, indicating to him that a
goal has been scored. Fig. 6a and 6b show the working of the
GoalControl and Hawk-Eye GLT systems, respectively.
The two systems promise that as long as 25% of the ball is
visible and at least two cameras are able to detect the ball, a
decision can be made.
VI. SUMMARY AND CONCLUSION
In this paper, we discuss the motivation in bringing
technology to the world of soccer. Eliminating human error is
the driving force behind goal-line technology systems today.
However, the systems are not perfect yet: while they are very
efficient at detecting a goal event up to millimeter levels, the
biggest hindrance is the exorbitant costs involved in setting
up and maintaining these elaborate systems.
We then explain a proposed model of a general goal-line
technology system, and provide an overview of the image
processing concepts involved in the working of a real-time
GLT system. Finally, we discuss two GLT systems that have
gained acceptance and have been used in tournaments at the
highest level of the game.
Like any piece of technological innovation, GLT succeeds in
simplifying a problem by automating tasks that are difficult
and, at times, impossible for the human mind to process. With
further improvements in the field of goal-line technology, we
can expect reduced costs, less ambiguity and more accuracy
when making decisions.
A video presentation, accompanied by a Python program to
demonstrate goal-line technology, can be found here.
VII. REFERENCES
[1] Wikipedia entry for Goal-Line Technology
(http://en.wikipedia.org/wiki/Goal_Line_Technology)
[2] Wikipedia entries for Hawk-Eye (https://en.wikipedia.org/wiki/Hawk-Eye)
and GoalControl (https://en.wikipedia.org/wiki/GoalControl)
[3] FIFAOfficial Website (http://www.fifa.com)
[4] FIFAQuality Program for Goal-Line Technology,Testing Manual 2014.
(http://quality.fifa.com/contentassets/148295d585194b1fb6030bbcd890b6
52/testing-manual-2014---fifa-quality-programme-for-goal-line-
technology-.pdf)
[5] Hawk-Eye Official Website (http://www.hawkeyeinnovations.com)
[6] GoalControl Official Website (http://www.goalcontrol.de/en)
[7] Kapadia, A., and Chimalapati, S. “Ball and Player Tracking in Sports.”
(http://www.ces.clemson.edu/~akapadi/research/ECE632_tracking.pdf)
[8] T. D’Orazio, M. Leo, “A review of vision-based systems for soccer video
analysis,” Elsevier Pattern Recognition 43 (2010) 2911–2926.
[9] Prayag Shah, Rishikesh Muchhala and Gaurang Shah, “A Review Paper
on Goal-Line Technology,” International Journal of Current Engineering
and Technology, Vol. 4, No. 5 (Oct 2014).
[10] T. D’Orazio, M. Leo, P. Spagnolo, M. Nitti, N. Mosca, A. Distante, “A
visual system for real time detection of goal events during soccer matches,”
Elsevier Computer Vision and Image Understanding 113 (2009) 622–
632.
[11] V.Pallavi a, Jayanta Mukherjee, Arun K. Majumdar, Shamik Sural, “Ball
detection from broadcast soccer videos using static and dynamic features,”
Elsevier J. Vis. Commun. Image R. 19 (2008) 426–436.
[12] Jens Christian Christensen Jensen, Gerth Brodal, “Event detection in soccer
[13] using spatio-temporal data,” Dept. of Computer Science, Aarhus
University, Nov. 2015.
[14] Yang Liu, Dawei Liang, Qingming Huang, Wen Gao, “Extracting 3D
information from broadcast soccer video,” Elsevier Image and Vision
Computing 24 (2006) 1146–1162.
[15] Paolo Spagnolo, Marco Leo, Pier Luigi Mazzeo, Massimiliano Nitti, Ettore
Stella, Arcangelo Distante, “Non-Invasive Soccer Goal Line Technology: A
Real Case Study,” in 2013 IEEE Conference on Computer Vision and
Pattern Recognition Workshops.
5
Fig. 6a: Working of GoalControl's GLT system
Fig. 6b: Working of the Hawk-Eye GLT system