Development of techniques for automatic detection of soccer 7 highlights

• Grant number: 16/15497-6
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: September 01, 2017
• End date: August 31, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Pascual Jovino Figueroa Rivero
• Grantee: Pascual Jovino Figueroa Rivero
• Company: Emanuel Thales Lara Piza
• City: Campinas
• Associated researchers: Pablo Alejandro Fonseca Arroyo

Abstract

Technology has an essential role to the development of any society: it makes resources more available to the masses. Such trend has caused image capture devices to be ever more accessible and abundant. Thus, there is increasing demand to solutions that deal with the massive amount of data raised. That same trend is impacting the sports' world too. Therefore, applications have to be developed in order to add value; i.e. monitoring and tracking; statistical analysis of videos, etc. However, it is hard to deal and resolve these complex problems even for professional sports teams, due to the highly specialization that is needed for handling properly all this big-data. The proposed research is focus on the soccer fields as a starting-point market, given the size of its market in Brazil. Our goal is to improve the experience and involvement of the players in this sport. Thus, fixed cameras are installed to automatically record the matches, to which are applied image processing techniques to automatically create content that will be presented back to the players, such as the highlights of the matches. The proposed service, of extraction and deployment of the soccer 7 players' highlights, has been done manually for a year, as can be seen in this link: http://tiny.cc/xroh8x published on the Esportes Company Youtube channel. The level of players' commitment indicates the potential for the creation of a sports social network, that will be fed continuously through the images captured in the fields. With the expansion of this services new features could be added, such as team and individuals statistics. To achieve all these goals; two important techniques of machine learning, image processing and deep learning will be used. The former is increasingly having relevant new scientific papers with great achievements, and the latter has already been tested many times by the scientific community. Having all these information in consideration, it is possible to provide an intelligence that would automatically generates relevant content to the players and also be flexible enough to be a scalable solution. (AU)