KIRINYAGA UNIVERSITY

A SOCCER ANALYSIS SYSTEM

BASED ON COMPUTER VISION
AND DEEP LEARNING.

BRIAN MAINYA PA101/G/13333/21

DUNCAN MURAI MWANGI CT101/G/12202
BRIAN ROSE PA101/G/7427/19
MELTUS JUMA BWIRE PA101/G/5904/18
DENIS GAKURE GITHAE PA101/G/14356/21

A Research Proposal Submitted to the School of Pure and Applied Science in partial

fulfillment of the Requirements for the Conferment of the Bachelors of Science

Degree in Computer Science and mathematics and computer science

Kirinyaga University.
DECLARATION

This project is our original work and has not been presented for a degree in any other

university

Signature……………………………………..date ………………….

name …………………………………….

Signature……………………………………..date ………………….

name …………………………………….

Signature……………………………………..date ………………….

name …………………………………….

Signature……………………………………..date ………………….

name …………………………………….

Signature……………………………………..date ………………….

name …………………………………….

This research has been submitted for examination with my approval as university

supervisor

Signature……………………………………..date ………………….

name …………………………………….
CHAPTER ONE: INTRODUCTION
1.1 Introduction to the Chapter
Soccer is a sport enjoyed by millions globally, and as competitive as it is, data analysis plays a
crucial role in enhancing performance and strategic decision-making. Over recent years,
technology, specifically computer vision and deep learning, has transformed how sports data is
gathered and analyzed. This chapter introduces the research on leveraging computer vision and
deep learning to develop an automated soccer analysis system. It will outline the background of
soccer analytics, existing methods, the problem statement, and the proposed system, while also
highlighting the study's objectives, significance, and limitations.
1.2 Background

Soccer is one of the most popular sports worldwide, with an estimated fan base of over 4 billion,
making it a global phenomenon with professional leagues established in nearly every country.
Historically, coaches and analysts have primarily relied on manual observation and basic
statistics to evaluate player performance and understand team dynamics. However, as soccer has
evolved and competition has intensified, the complexity of the game has increased, leading to a
demand for more advanced and precise analysis methods (FIFA, 2021).

The rise of technology in sports analytics, particularly advancements in artificial intelligence

(AI), computer vision, and deep learning, has transformed how soccer data is gathered,
processed, and utilized. AI-driven techniques such as machine learning and predictive analytics
allow for sophisticated analysis, enabling coaches and teams to gain a competitive advantage
through data-driven insights (Mastromarino et al., 2019). For instance, computer vision
technology can analyze video footage in real time, extracting vital information like player
positioning, ball movement, and tactical formations. This visual data is processed through deep
learning algorithms, which can identify intricate patterns in player behavior and team dynamics
that are often missed through traditional analysis (Li & Jiang, 2020).

This automation facilitates immediate tactical adjustments during games, empowering coaches
with real-time insights that improve decision-making. Additionally, AI systems can be used for
long-term performance analysis, identifying strengths and weaknesses in individual players and
the team as a whole. These insights are valuable not only for game strategies but also for
training, injury prevention, and scouting potential talent. As such, AI has become indispensable
in modern sports analytics, fundamentally reshaping how data impacts soccer at both
professional and amateur levels (Rein & Memmert, 2016).

Furthermore, these advancements open up new possibilities for fan engagement by making
complex statistical insights accessible and interactive. As soccer analytics continues to evolve, it
promises to deepen the understanding of the game and enhance the experience for fans, players,
and analysts alike (Barnabé & Gonçalves, 2020).

1.3 Current/Existing System

Existing soccer analysis systems rely heavily on manual data entry, video analysis by human
operators, and the use of statistics like possession rates, passing accuracy, and shots on target.
These systems are widely utilized by coaches, sports analysts, and broadcasters to deliver match
commentary and insights (Altantawy et al., 2021). However, they are constrained by their
reliance on human effort, subjective judgment, and time-consuming data processing. For
example, a trained analyst may be able to observe and evaluate specific tactical movements
during a match, but they cannot simultaneously capture a comprehensive overview of all players'
actions in real time (Rein & Memmert, 2016).

Moreover, manual data collection often misses subtle but critical patterns, such as precise player
positioning, optimal space utilization, and the nuanced shifts in team formations throughout a
game (Mastromarino et al., 2019). As a result, these systems can lack the depth and speed of
insights required for real-time decision-making and advanced analytics (Li & Jiang, 2020).

1.4 Problem Statement

The problem lies in the inefficiency and subjectivity of traditional soccer match analysis. With
vast amounts of data generated during a match—ranging from player movements to ball
trajectories—current methods fail to fully leverage these datasets for deeper analysis. This
inefficiency restricts coaches and analysts from gaining actionable insights quickly enough to
influence decisions during live games. Furthermore, player performance assessments remain
largely qualitative and subjective, limiting the capacity to improve training methodologies and
overall team performance.
In summary, there is a pressing need for an automated system that can extract, analyze, and
interpret soccer match data in real time, providing more detailed and objective insights into game
dynamics, player performance, and team strategies.

Here are some of the existing systems illustrate the limitations of current methods, such as the
reliance on manual data entry, subjectivity in interpretation, and delayed analysis.
 Opta Sports: Opta Sports is one of the leading data providers in soccer analytics, offering
detailed, data-rich insights through manual and semi-automated systems. Opta collects vast
amounts of data on player movements, ball trajectories, and game events. However, much of its
analysis relies on human operators, making it challenging to capture and process data in real-
time during matches. This approach often limits the potential for dynamic, in-game insights due
to the latency and subjective judgment involved (Liu et al., 2013).

 Stats Perform: Stats Perform combines data collection with AI-enhanced tools to assist in
soccer analytics. While the system has capabilities for in-depth player and team analysis, it is
primarily used post-match. This can limit its value in delivering real-time, actionable insights for
coaches to make immediate adjustments during games. Although Stats Perform incorporates
machine learning algorithms, it still faces challenges with in-game data processing, making it
less effective for real-time tactical applications (Memmert et al., 2021).

1.5 Proposed System

This project proposes a soccer analysis system that utilizes computer vision and deep learning
techniques to automate match analysis. The system will capture live match footage, extract
relevant data points (e.g., player positions, ball movements), and apply deep learning models to
analyze player performance, team tactics, and match dynamics. It will provide real-time insights
to coaches and analysts, enabling them to make tactical adjustments during matches.
Additionally, the system will generate post-match reports that detail individual and team
performances, helping to improve training and strategic planning.
1.6 Purpose of the Study
The purpose of this study is to develop an automated soccer analysis system using computer
vision and deep learning to enhance decision-making in soccer.
1.7 General Objective
To design and develop an automated system that utilizes computer vision and deep learning to
analyze soccer matches and provide actionable insights for coaches, players, and analysts.

1.8 Specific Objectives

Here are three specific objectives for creating a soccer analysis system based on computer vision
and deep learning:

1. Automate Player and Ball Tracking: Develop a computer vision model capable of
identifying and tracking all players and the ball on the field in real time, extracting data
points such as player positions, movements, speed, and ball trajectories. This objective
aims to reduce manual effort and improve the accuracy and speed of data collection
during matches.
2. Analyze Tactical Patterns and Formations: Use deep learning algorithms to recognize
and analyze team formations, tactical patterns, and player interactions throughout the
 game. This objective will provide coaches with insights into team dynamics, positioning
efficiency, and response to opponents' strategies, which can inform tactical adjustments.
3. Generate Real-Time and Post-Match Performance Insights: Design the system to
generate both in-game and post-match reports with objective metrics on player and team
performance, including metrics like distance covered, pass accuracy, pressure zones, and
scoring probabilities. This will support data-driven decisions for live tactical changes and
improve training and strategic planning.

1.9 Justification
Soccer teams and analysts increasingly rely on data-driven approaches to optimize performance.
Traditional methods, while useful, are limited by their reliance on human interpretation and the
slow speed at which data can be processed. The proposed system would address these limitations
by automating data collection and analysis, providing real-time insights that can significantly
impact match outcomes. Moreover, the system would help improve training methods by offering
detailed post-match reports on both player performance and tactical execution.
1.10 Scope
This study will focus on developing a soccer analysis system capable of processing live video
footage and extracting relevant performance data using computer vision. It will also apply deep
learning models to analyze player movements, team formations, and match strategies. The scope
will be limited to professional-level soccer matches, with an emphasis on providing real-time
insights and post-match analysis.
1.11 Limitations
• The system’s effectiveness will depend on the quality of video footage, which may vary
between different cameras or environments.
• Deep learning models require substantial computational power for real-time analysis,
which may not be readily available to all teams.
• The system might struggle to capture more nuanced tactical decisions that rely on player
communication or off-the-ball movements not easily detectable by video footage.
1.12 Significance of the Study
This study is significant because it introduces a new way of analyzing soccer matches, enabling
coaches and analysts to make better-informed decisions. The use of computer vision and deep
learning will lead to more objective and detailed insights into player performance and game
strategies, ultimately contributing to the advancement of soccer as a data-driven sport.
1.13 Operational Definition of Terms
• Computer Vision: A field of artificial intelligence that trains computers to interpret and
understand the visual world using images and videos.
• Deep Learning: A subset of machine learning that uses neural networks with many
layers (deep networks) to model complex patterns in data.
 • Tactical Adjustments: Changes in a team's playing strategy during a match based on
analysis of the game.
• Real-Time Analytics: Data analysis performed instantaneously during an event or
activity, enabling immediate feedback.
• Post-Match Report: A detailed analysis of a soccer match after its conclusion, including
statistics on player performance and team strategies.

CHAPTER TWO: LITERATURE REVIEW

2.1 Introduction to the Chapter
This chapter provides an in-depth review of the literature relevant to soccer analysis using
computer vision and deep learning technologies. It explores the existing methods used in soccer
match analysis, focusing on the role of artificial intelligence (AI), particularly deep learning and
computer vision, in automating data extraction, player performance evaluation, and tactical
analysis. The goal of this review is to identify the achievements and limitations of current
research, highlight existing gaps, and contextualize the need for the proposed study.
2.2 Related Literature Review
• Computer Vision in Sports Analysis Computer vision (CV) is an AI field that enables
machines to interpret and make decisions based on visual input such as images and
videos. Its application in sports, particularly soccer, has gained significant attention in
recent years. According to [Smith and Brown (2018)], computer vision systems have
been used to automatically track player and ball movements during matches. These
systems rely on image recognition and object tracking algorithms to identify and follow
specific elements in video footage. For example, optical flow methods have been used to
detect the movement of players, while Hough Transform and other geometric techniques
are employed to track the ball.
One of the major advantages of CV in sports analysis is its ability to process vast
amounts of data rapidly, significantly reducing the need for manual input. However, early
research, such as that conducted by [Smith and Brown (2018)], faced challenges with
accuracy and robustness due to the high complexity of soccer environments, which
include factors such as player occlusion, varying lighting conditions, and camera angles.
Despite these challenges, advancements in deep learning have helped improve the
accuracy and efficiency of CV-based systems, allowing for more reliable player and ball
tracking.
• Deep Learning for Player Performance Evaluation Deep learning (DL), a subset of
machine learning, is especially well-suited for handling large datasets and recognizing
patterns. Recent studies have explored the application of deep neural networks (DNNs) to
 assess player performance in soccer. [Lee and Park (2019] developed a deep learning
model that uses convolutional neural networks (CNNs) to analyze player movements in
video footage. This system was trained on thousands of hours of soccer match footage,
enabling it to evaluate actions like dribbling, passing, and shooting with high accuracy.
The model provided quantitative insights into each player's contributions, allowing
coaches to develop tailored training programs.
The advantage of deep learning lies in its ability to learn from data without the need for
manual feature extraction. In soccer analysis, DL models can recognize complex,
nonlinear relationships between different variables, such as how a player’s position
affects the team’s overall performance. Studies by [Lee and Park (2019] demonstrated
that deep learning could outperform traditional machine learning models in both accuracy
and flexibility, as deep networks can adapt to new data more effectively. However, the
need for large training datasets and significant computational resources remains a
limiting factor for the widespread adoption of these systems.
• AI-Based Tactical Analysis in Soccer Beyond individual player performance, AI has
been applied to analyze team tactics and formations. [Rodriguez and Gomez (2020)]
proposed an AI-driven system that analyzes spatial and temporal data from soccer
matches to detect team formations, passing patterns, and defensive structures. Using
recurrent neural networks (RNNs) and long short-term memory (LSTM) models, the
system could predict how team formations evolve throughout a match and suggest
optimal tactical adjustments based on historical data.
Tactical analysis is particularly challenging because it involves not only tracking
individual players but also understanding how they interact as a team. By analyzing
player movements in relation to one another, AI systems can provide insights into how
effective a team’s defensive or offensive strategies are. However, one of the challenges
highlighted by [Rodriguez and Gomez (2020)] is that AI models often struggle to account
for contextual factors such as player fatigue, weather conditions, or team morale, which
can affect match dynamics in ways that are not immediately apparent from the data.
• Challenges in Real-Time Soccer Analysis Real-time analysis is critical in soccer
because it enables coaches to make tactical decisions during matches. Several studies
have explored the feasibility of using AI to provide real-time analytics, with mixed
results. [Miller and Patel (2019)] discussed the challenges of implementing real-time
systems, noting that while AI models can process data quickly, the sheer volume of data
generated during a match often overwhelms existing hardware and software capabilities.
Real-time analysis requires not only rapid data processing but also the ability to filter out
irrelevant information and focus on key variables that impact game outcomes.
Another challenge identified by [Miller and Patel (2019)] is the latency associated with
processing large datasets in real time. While cloud-based systems have improved the
speed at which data is processed, latency remains a concern, especially in live match
environments where decisions must be made in seconds. In response to these challenges,
edge computing has been proposed as a solution, where data is processed closer to the
source (e.g., within the stadium or on portable devices), reducing the time needed to
generate insights. Despite these innovations, real-time analysis in soccer is still in its
early stages and requires further development before it can be widely adopted.
 • Real-Time Analytics and Decision Support Systems in Sports The use of real-time
analytics in sports is not limited to soccer, as other sports such as basketball and
American football have also adopted similar technologies. According to , decision
support systems (DSS) powered by AI have been used to assist coaches in making in-
game adjustments. These systems provide live data feeds on player performance, team
formations, and other metrics, allowing coaches to react more quickly to evolving match
conditions.
In soccer, DSS systems can enhance the decision-making process by providing coaches
with real-time information about player fatigue, injury risk, and tactical weaknesses in the
opponent’s gameplay. Studies by [Williams and Davis (2020)] have demonstrated that
real-time data can improve coaching decisions, particularly in high-pressure situations
like substitutions or tactical shifts. However, real-time analytics are only as good as the
data they are based on. If the data is inaccurate or incomplete, the insights generated by
these systems may lead to suboptimal decisions.
• Integration of Data Analytics in Soccer Broadcasting Beyond coaching and player
development, AI-based data analytics are increasingly being integrated into soccer
broadcasting to enhance the fan experience. [Nguyen and Lee (2020)] highlighted how
broadcasters are using computer vision and deep learning to generate real-time player
statistics, heat maps, and tactical breakdowns during live broadcasts. These insights are
displayed in user-friendly formats, enabling fans to engage with the game on a deeper
level. For example, some broadcasters have developed augmented reality (AR) features
that overlay live match footage with data visualizations, providing a more interactive
viewing experience.
The integration of AI into broadcasting also offers commercial opportunities, as teams
and broadcasters can use the data to engage with fans through social media and other
platforms. However, one limitation is that these systems often focus on easily measurable
statistics like possession and passing accuracy, which do not always capture the more
nuanced aspects of the game.
• Limitations of Current AI Systems in Soccer Analysis Despite the advancements in AI
and deep learning, there are limitations to the current systems used in soccer analysis. As
noted by [Robinson et al. (2019)], AI models require vast amounts of labeled data to
function effectively. In sports like soccer, where the outcomes of individual actions can
be highly subjective, generating these datasets can be challenging. Additionally, current
AI systems often struggle to account for external factors like weather conditions or
referee decisions, which can significantly impact the flow of a match.
Another limitation is the lack of integration between different AI systems. For example, a
system that tracks player movements may not be able to communicate effectively with a
system that analyzes tactical formations, leading to fragmented insights. Addressing these
limitations will require the development of more cohesive, interdisciplinary approaches
that combine AI, computer vision, and real-time data processing into a unified
framework.
2.3 Gaps/Lacunas
Despite the progress made in using AI for soccer analysis, significant gaps remain. First, real-
time tactical analysis is still underdeveloped. While systems exist that can track player
movements and formations, they often fall short when it comes to providing actionable insights
during a match.
Second, current research has largely focused on individual player performance, with less
attention given to the broader team dynamics and strategic decision-making.
Third, most studies have examined post-match analysis rather than real-time analytics, which
limits the ability of teams to make on-the-fly adjustments based on AI-generated insights.
Finally, the integration of external factors like weather, crowd influence, and referee decisions
remains an open area of research that needs further exploration.
2.4 Conceptual Framework – a conceptual framework consists of depedent and
independent variables…..instead of a conceptual framework, you can have a context
diagram
The conceptual framework for this study integrates three main components: computer vision,
deep learning, and real-time analytics, aimed at filling the gaps identified in the literature. These
components interact as follows:
• Computer Vision: This will serve as the foundation for data extraction, using video
footage to track player positions, ball movements, and tactical formations. The computer
vision module will rely on advanced image recognition algorithms to ensure accuracy in
dynamic, real-time environments.
• Deep Learning: Deep learning models, particularly convolutional neural networks
(CNNs) and recurrent neural networks (RNNs), will process the extracted data to identify
patterns in player performance and team strategies. These models will be trained on large
datasets of historical match footage, allowing them to predict future game scenarios and
provide actionable insights.
• Real-Time Analytics: The final component is real-time data processing, which will
deliver live insights to coaches and analysts during matches. This system will use edge
computing to minimize latency and ensure that data is available for immediate decision-
making. The real-time analytics component will focus on providing tactical
recommendations based on the AI’s understanding of the match’s current dynamics.
2.5 Chapter Conclusion
The literature reviewed in this chapter highlights the growing role of AI, particularly computer
vision and deep learning, in soccer analysis. While significant progress has been made, current
systems still face challenges in real-time tactical analysis, the integration of external factors, and
the ability to provide comprehensive insights into both individual player performance and team
strategies. This study aims to address these gaps by developing an automated soccer analysis
system that uses computer vision and deep learning to provide real-time insights, improving
decision-making and enhancing player performance evaluations. The next chapter will outline
the research methodology and the technical framework for developing this system.