Video analysis for replay detection in sport events

The postproduction cost of a sport event video requires lots of resources dedication and expenses of time trying to find the best highlights moments that will be used, for instance, in creating the summary of the event. This process can be optimized and improved in efficiency. During the event, the...

ver descrição completa

Detalhes bibliográficos
Autor: Martínez Junyent, David
Formato: tesis de maestría
Fecha de publicación:2012
País:España
Recursos:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2099.1/15505
Acesso em linha:https://hdl.handle.net/2099.1/15505
Access Level:acceso abierto
Palavra-chave:Image processing
Speech processing systems
Descriptors audiovisuals
Classificació d'events
Processament de imatge
Processament d'àudio
Imatge -- Processament -- Tècniques digital
Audiovisuals
Vídeo digital
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Processament del senyal::Processament de la imatge i del senyal vídeo
Descrição
Resumo:The postproduction cost of a sport event video requires lots of resources dedication and expenses of time trying to find the best highlights moments that will be used, for instance, in creating the summary of the event. This process can be optimized and improved in efficiency. During the event, the most important moments are repeated to offer to the audience the outstanding scene several times and from different points of view. The objective of the project is to automatically find the replays in live or pre-recorded transmission and accelerating the post-production process. The results will be part of the project CENIT-E BUSCAMEDIA CEN20091026, developed in the studios of Televisió de Catalunya (TVC) and which are focused on automated generation through content analysis. A software has been developed to detect the replays for different kind of sport events, principally soccer. This, implements many operation modes detailed during this report. We find from a mode rather manual to a full automatic mode, and moreover the percentages of success are presented after testing then using some videos from the TVC database. The structure of the work has been divided into five major sections: The first chapter begins by introducing us to the context in which it places the project, proposing the objectives to be achieved, and also discusses the data and tools used for their development. Subsequently, there is exposed the state of the art with a collection of methods used for the detection of repeats, which are the foundations on which we developed our methodology. The third chapter is the longest and complex. This contains the entire process of experimentation and improvements planned from the inception until the system implemented. In addition, the following section talks about the technical and exhibits the algorithm implemented in form of block diagram detailing all the operation modes. Finally, the last chapter contains all the results and conclusions after applying the algorithm on a set of videos taken from the database o f TVC, as well as its application in other areas such as Formula1 videos.