Soccer Players Real Location Determination using Perspective Transformation

Stephen Karungaru, Hiroki Tanioka, Kenji Matsuura

Abstract: Recently, the use of Multi-Object Tracking(MOT) technology has become vital for sports analysis using video. The main advantages over GPS are mainly cost and non-interference with the players. However, to estimate the same real location comparable to the GPS, camera calibration and play area dimensions detection are necessary. This work, proposes a deep learning method for player, ball and field markers detection and the use of the perspective transform to determine the real location of the players in the field irrespective of the camera position. Our results show that, this technology could be useful and cheap for both amateur and professional teams.