People often make mistakes, that is why we try to automate every aspect of our lives. The field of sports is not an exception. While only a bit more than a decade ago people analyzed games, this is now done by artificial intelligence. Because of its fast development in the last ten years, neural networks are now faster, more accurate and in those metrics better then its human counterparts in some fields. The motivation for the master's thesis is thus to develop an algorithm, that can detect player statistics during an NBA broadcast. It would also help the user to better understand the game with the use of augmented reality. The aim was to create an algorithm that could detect the players on the court and track their actions with the highest accuracy possible. In the framework of the master's thesis we studied modern and effective methods in the knowledge domain. We developed an algorithm that could successfully detect players on the court, and classify them to their respective team with a neural network. With the use of a the homography transformation we moved the positions of the players to a two dimensional space on the court. We defined a new algorithm to detect the actions of the players, and thus their statistics. During the implementation of the algorithm we tried different methods to solve the problem. We analyzed their effectiveness and discussed their strengths and weaknesses.
|
