Anna Szilagyi/Epa
Diogo Costa is the “king of penalties”, but AI debates with the National Team goalkeeper
Positive note for machines, negative for human guardians. Artificial intelligence surpasses goalkeepers to predict that penalties are beaten, with an effectiveness of 64% between left and right.
A model of deep learning He was able to predict more effectively than human goalkeeper to which side a penalty would be beaten, according to a new Arxiv study that analyzed over a thousand great penalties.
Researchers at the University of Las Palmas de Gran Canaria in Spain, led by David Freire-Abobón, analyzed more than a thousand royal gambling penalties transmitted on television to realize whether artificial intelligence (AI) could overcome goalkeeper intuition by predicting the direction of the rims based on players’ body language.
Of the 1010 penalties collected, 640 were considered appropriate for automatic analysis. These videos were analyzed by 22 different models of deep learningwho tried to predict if the shot would go to the left, right or center of the goal. It is important to emphasize that the filming and the fact that the player is left -handed or right -handed were only considered.
The best of the models was able to properly predict the direction of the shot in 52% of cases – slightly above the 46% successful attributed to the real goalkeeper. When researchers eliminated the “medium” option of the analysis, accuracy rose to 64%according to.
“It’s impressive how small body clues, even before the foot touching the ball, can reveal intentions,” said study author Freire-Abregón.
Despite the good results, the researcher admits that applying these predictions in the context of the game is still challenging. The next step will be to evaluate whether it is possible to anticipate the penalties’ direction only based on the initial movements of players – and in advance these predictions can be made without losing effectiveness.
