Reconhecimento automático de pose empregado na estimação da posição da bola em partidas de futsal

Abstract

Determining the location of the ball in futsal matches is of great importance for team technical reports, sports broadcasts and any other pertinent analysis. However, this recognition does not take place in a trivial way, due to its small size, the great speed it can acquire and also the fact that it is obstructed by players, it becomes a difficult task for conventional computer vision recognition methods or doing the use of numerous equipment. A plausible resource to be used, but hardly used for this demand, is the pose of the players during the matches. This project performs the acquisition of these poses as well as evaluating a relation between them and the original position of the ball, thus employing regression algorithms to predict its position. Performing this estimation in a generic way independent of camera or recording location.