Unified Player Identification Across Field Sports Survey And Proposed Model

Abstract

With the popularity of sports industry and development of rapid technological innovations, the player detection has become an important task in sport images/video analysis. AI is now widely used for game prediction, performance analysis, highlight generation, and assistant coaching to improve team performance and decision-making. If the players have been detected accurately, this information can then be effectively used to perform high level analysis. The interest of this paper is to analyze different approaches used to identify and detect the players in sports images proposed and develop a strong player detection system to detect, recognize and track them. The conventional techniques work by detecting the face or body of the players/humans in the picture and then applying state-of-the-art face recognition methodologies to identify the player. Then there are popular systems that work by recognizing the player number on the jersey of players or the bib of the athletes. The approaches have been developed to track the players based on their locations in the field and finally detect them. This study shows the comparative analysis along with the pros and cons of the methodologies discussed in detail. To the best of our knowledge, such a detailed review of player detection comparing various approaches is unavailable in the literature and thus, our contribution claims are significant. A simple and robust automatic technique is proposed to estimate the location of the racing bib. To fulfill the task of athlete identification we have made use of the bib number as well as the facial features. While doing the identification using bib number, facial features are also extracted and a database is built for identification by recognizing face. Extensive test findings are displayed on actual marathon photos with varied player counts in varied positions and illumination settings. It achieves state-of the-art results up to 97%.