Thesis SISTEMA DE VISIÓN POR COMPUTADOR PARA DETECCIÓN Y SEGUIMIENTO DE BALÓN DE FUTBOL EN CASOS COMPLEJOS UTILIZANDO CÁMARAS ESTÁTICAS
Date
2017
Authors
CÁRDENAS NAHUEL, DANIEL GERMÁN
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Este trabajo forma parte de un proyecto el cual busca implementar un sistemade video-analisis en partidos competitivos, con el proposito de entregar informacion estadística a cuerpos tecnicos para dar soporte a la toma de decisiones.Estos sistemas utilizan camaras estaticas alrededor del estadio para extraer informacion relacionada a el desempeño de los jugadores y sus interacciones.En particular, esta tesis se centra en el caso del balon de futbol, el cual es un elemento crucial para el desarrollo del juego, ya que este permite obtenerinformacion relevante para el entendimiento de los distintos acontecimientos del partido. El seguimiento automatico de balon presenta varios problemasprincipalmente debido a su pequeno tamaño en el video, el cual dificulta la extraccion de sus caracteristicas y lo hace mas sensible a distorcionarse cuandopresenta una alta velocidad. Ademas, existen problemas cuando el balon interactúa con los jugadores, porque este presenta ocultaciones totales o parciales,o su dinamica se vuelve impredecible.En este trabajo incorpora soluciones del estado del arte, agregando a estosuna representacion elptica del balon, la cual hace mas rapido y certerosu reconocimiento. Por otro lado, tambien incorpora un robusto sistema deseguimiento de balon para casos de ocultacion.Esta investigacion se enmarca dentro del proyecto FONDECYT 11121383:"Methodology and applications for incremental Behaviour Learning in Videoguided by information Reliability".
This work is part of a project which implements a video-analysis system, withthe purpose of bringing statistical information to support technical sta indecision making for competitive matches in collective sports. These systemsuse static cameras installed around the stadium and utilise their captures toextract relevant information relative to the performance of players and theirinteractions.In particular, this thesis focuses on the soccer ball, a crucial element in thegame, because it presents a relevant information for the understanding of thedynamics of gameplay. Automatic ball tracking presents several challenges,mainly because of the small size of the ball in the video, which dicults theextraction of its features and makes it more sensitive to distortion when theball is in high speed. Moreover, there are problems when the ball interactswith players, because it presents partial or total occlusion, or its dynamicsbecome unpredictable.This work utilises state-of-the-art algorithms, incorporating an ellipticalrepresentation into the ball model, which makes its detection more reliable andfast. Besides, a robust tracking system is incorporated to deal with occlusionsituations.This research has been supported, in part, by Fondecyt Project 11121383:"Methodology and applications for incremental Behaviour Learning in Videoguided by information Reliability".
This work is part of a project which implements a video-analysis system, withthe purpose of bringing statistical information to support technical sta indecision making for competitive matches in collective sports. These systemsuse static cameras installed around the stadium and utilise their captures toextract relevant information relative to the performance of players and theirinteractions.In particular, this thesis focuses on the soccer ball, a crucial element in thegame, because it presents a relevant information for the understanding of thedynamics of gameplay. Automatic ball tracking presents several challenges,mainly because of the small size of the ball in the video, which dicults theextraction of its features and makes it more sensitive to distortion when theball is in high speed. Moreover, there are problems when the ball interactswith players, because it presents partial or total occlusion, or its dynamicsbecome unpredictable.This work utilises state-of-the-art algorithms, incorporating an ellipticalrepresentation into the ball model, which makes its detection more reliable andfast. Besides, a robust tracking system is incorporated to deal with occlusionsituations.This research has been supported, in part, by Fondecyt Project 11121383:"Methodology and applications for incremental Behaviour Learning in Videoguided by information Reliability".
Description
Catalogado desde la version PDF de la tesis.
Keywords
DETECCION DE OBJETOS , EXTRACCION DE CARACTERISTICAS , SEGUIMIENTO DE OBJETOS , VISION POR COMPUTADOR
