| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 20.08 MB | Adobe PDF |
Authors
Abstract(s)
A evolução da tecnologia trouxe uma alargada gama de aplicações informáticas em inúmeras
áreas. Uma delas é o desporto, e em particular o futebol. Neste momento, a tecnologia
desempenha um papel fundamental para a melhoria do desempenho e do planeamento de
táticas de jogo, fornecendo informações descritivas que ajudam atletas e treinadores a tomar
decisões informadas. No entanto, as instituições amadoras e de formação, que normalmente
não dispõem de grandes recursos financeiros e tecnológicos, enfrentam desafios na automatização
dos processos da extração de dados relacionados com os jogos em contraponto com
as entidades profissionais que possuem meios próprios de aquisição de ferramentas tecnológicas
de suporte à deteção de eventos e análise de jogos.
Este projeto visa não apenas colmatar uma lacuna tecnológica nas escolas de formação e
nos clubes amadores, mas também apoiar o seu desenvolvimento, proporcionando uma ferramenta
acessível para a deteção automática de eventos durante os jogos de futebol.
Esta dissertação explora essa lacuna através da aplicação de técnicas de Inteligência Artificial
e Visão Computacional na análise de jogos de futebol através da utilização de vídeos
dos mesmos.
A transformação do conteúdo de vídeos de jogos em dados e estatísticas necessita de um
grande esforço humano e depende de serviços dispendiosos, o que dificulta o seu acesso a
instituições amadoras e de formação. Essa transformação pode também ser utilizada por
aplicações de infromáticas que disponibilizam resultados de futebol e estatísticas de jogos
de futebol permitindo aos adeptos dos clubes um acesso aos dados recolhidos, aproximando
assim os apoiantes às suas equipas. A dificuldade de acesso a essa informação nestas organizações
cria um fosso entre os clubes profissionais e os amadores. O trabalho desenvolvido e
descrito neste documento tem como principal objetivo conceber um mecanismo de deteção
e análise de eventos em suporte de vídeo, em tempo real, aplicados a jogos de futebol com
recurso a técnicas da Inteligência Artificia e da Visão Computacional.
Este projeto consistiu na realização de um algoritmo que permite detetar golos e outros
eventos possibilitando a interpretação de um jogo de futebol através de suporte de vídeo do
mesmo. Na tentativa de deteção desses eventos, foram testadas duas abordagens diferentes
ao problema, a deteção de objetos e a deteção de eventos. Através do estudo realizado
percebeu-se que a melhor abordagem seria a utilização de modelos de deteção de eventos.
Foram, assim, aplicados dois tipos de modelos de deteção de eventos: Convolutional Long
Short-Term Memory (ConvLSTM) e Long Recurrent Convolutional Network (LRCN).
Um dos requisitos do algoritmo é mostrar-se adaptável às condições de jogo e conseguir promover
uma melhoria na análise do desempenho das equipas e jogadores e no envolvimento
dos adeptos com as instituições amadoras e de formação.
O protótipo desenvolvido foi capaz de detetar golos, cantos e penalties. Em termos de
resultados os golos apresentaram os melhores resultados, embora com uma precisão relativamente
baixa. No entanto, foi possível identificar o modelo mais adequado para este tipo
de desafio.
The evolution of technology has brought a wide range of computer applications to countless areas. One of these is sport, and football in particular. At the moment, technology plays a role in improving performance and planning game tactics, providing descriptive information that helps players and coaches make informed decisions. However amateur and training institutions, which usually don’t have large financial and technological resources, face challenges in automating the processes of extracting data related to matches, in contrast to professional organisations that have their own means to acquire technological tools to support event detection and match analysis. games. This project aims not only to fill a technological gap in training schools and amateur clubs, but also to support their development by providing an affordable tool for the automatic detection of events during football matches. events during football matches. This dissertation explores this gap by applying Artificial Intelligence and Computer Vision techniques to analyse football matches using match videos. Transforming match video content into data and statistics requires a great deal of human effort and depends on expensive services, which makes it difficult for amateur and training institutions to access. This transformation can also be used by infromatics applications that provide football results and match statistics, giving club supporters access to the data collected, thus bringing supporters closer to their teams. The difficulty of accessing this information in these organisations creates a gap between professional and amateur clubs. The main aim of the work carried out and described in this document is to design a mechanism to mechanism for detecting and analysing real-time video events applied to football matches using Artificial Intelligence and Computer Vision techniques. This project consisted of creating an algorithm to detect goals and other events, making it possible to interpret a football match using video support. In an attempt to detect these two different approaches to the problem were tested: object detection and event detection. Through the study carried out, it was realised that the best approach would be to use event detection models. Two types of event detection models were therefore applied: Convolutional Long Short-Term Memory (ConvLSTM) and Long Recurrent Convolutional Network (LRCN). One of the algorithm’s requirements is to be adaptable to match conditions and to be able to promote an improvement in analysing the performance of teams and players and the involvement of fans with amateur and training institutions. The prototype developed was able to detect goals, corners and penalties. In terms of results, goals showed the best results, albeit with relatively low accuracy. accuracy. However, it was possible to identify the most suitable model for this type of challenge.
The evolution of technology has brought a wide range of computer applications to countless areas. One of these is sport, and football in particular. At the moment, technology plays a role in improving performance and planning game tactics, providing descriptive information that helps players and coaches make informed decisions. However amateur and training institutions, which usually don’t have large financial and technological resources, face challenges in automating the processes of extracting data related to matches, in contrast to professional organisations that have their own means to acquire technological tools to support event detection and match analysis. games. This project aims not only to fill a technological gap in training schools and amateur clubs, but also to support their development by providing an affordable tool for the automatic detection of events during football matches. events during football matches. This dissertation explores this gap by applying Artificial Intelligence and Computer Vision techniques to analyse football matches using match videos. Transforming match video content into data and statistics requires a great deal of human effort and depends on expensive services, which makes it difficult for amateur and training institutions to access. This transformation can also be used by infromatics applications that provide football results and match statistics, giving club supporters access to the data collected, thus bringing supporters closer to their teams. The difficulty of accessing this information in these organisations creates a gap between professional and amateur clubs. The main aim of the work carried out and described in this document is to design a mechanism to mechanism for detecting and analysing real-time video events applied to football matches using Artificial Intelligence and Computer Vision techniques. This project consisted of creating an algorithm to detect goals and other events, making it possible to interpret a football match using video support. In an attempt to detect these two different approaches to the problem were tested: object detection and event detection. Through the study carried out, it was realised that the best approach would be to use event detection models. Two types of event detection models were therefore applied: Convolutional Long Short-Term Memory (ConvLSTM) and Long Recurrent Convolutional Network (LRCN). One of the algorithm’s requirements is to be adaptable to match conditions and to be able to promote an improvement in analysing the performance of teams and players and the involvement of fans with amateur and training institutions. The prototype developed was able to detect goals, corners and penalties. In terms of results, goals showed the best results, albeit with relatively low accuracy. accuracy. However, it was possible to identify the most suitable model for this type of challenge.
Description
Keywords
Inteligência articial Visão computacional Futebol amador e de formação Deteção de eventos Automatização Artificial intelligence Computer vision Amateur and youth football Event detection Automation