Type of view estimation in Football sequences

Due to the huge repercussion of football broadcast in society, an enormous number of applications can be derived to both analyze the match and enhance the visual experience of the spectator. These applications request semantical information about the content of the images. In particular, the type of...

Descripción completa

Detalles Bibliográficos
Autor: Varas González, David
Tipo de recurso: tesis de maestría
Fecha de publicación:2010
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2099.1/21364
Acceso en línea:https://hdl.handle.net/2099.1/21364
Access Level:acceso abierto
Palabra clave:Images -- Classification
Image processing
classification
football
view
descriptor
decission tree
Imatges -- Classificació
Imatges -- Processament
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Processament del senyal::Processament de la imatge i del senyal vídeo
Descripción
Sumario:Due to the huge repercussion of football broadcast in society, an enormous number of applications can be derived to both analyze the match and enhance the visual experience of the spectator. These applications request semantical information about the content of the images. In particular, the type of view in a football image contains valuable information about the game. Thus, the type of view must be automatically computed to be able to process the large amount of information extracted from each football match. In this work, we propose a robust classification system that estimates the type of view in football images in real time. For each frame of the sequence, a set of descriptors is extracted to characterize a specific part of the scene: the grass field. Gathering all these descriptors and a few ones related with texture, a decision tree determines the view that is shown in that frame. In order to improve the robustness of the algorithm, the redundancy of the temporal domain is exploited. The validity of the proposed algorithm has been tested on a large amount of frames from broadcasted football sequences in a wide variety of scenarios (stadiums, light conditions, ...). Promising results have been obtained with a 96% of accuracy in the classification of these images.