Analyzing the preliminary movements, such as posture, running angle and inclination of the bust, the IA has foreseen the direction of a penalty with more accuracy of the goalkeepers.
Is artificial intelligence better than Donnarumma, Buffon and Zoff? Maybe yes.
A team led by David Freire-Obregón, of the University of Las Palmas de Gran Canaria (Spain), has developed models of Deep Learning capable of predicting the direction of a penalty (left, right or center) with a precision higher than that of the real goalkeepers.
Analyzing real videos of about six hundred penalties, the IA first learned the shooting techniques of various players, and then managed to beat human intuition. But how did he do it? Simply by analyzing the preliminary movements, such as posture, running angle and inclination of the bust, from which signs emerge that our brain struggles to grasp in real time.
Automatic prediction. In detail, the researchers fished between 1,010 clips of royal penalties pulled in various Spanish championships, discarding insufficient quality videos and selecting 640 high definition movies to be used to train the models. 22 different variants of the IA model (in technical jargon: “architecture”) have been experienced on these videos with the aim of predicting the direction of the shot before the player’s foot hit the ball. The best model obtained 52% of correctness in predating if the shot would have been pulled to the left, right or in the center, a result that beats the average of the goalkeepers, estimated at around 46%. When the “central shooting” option is removed from the calculation, however, the accuracy rises a lot, reaching 64% and offering a margin of 10 percentage points compared to human skill (which also rises, but only up to 54%). The researchers say that even minimal gestures before the shot, as a positioning of the foot and thin inclinations of the body, contain useful signals to predict the footballer’s intent.
Areas of use. The research was not obviously carried out for a simple statistical exercise, but with the aim of applying new knowledge to the training of goalkeepers, thus using the IA to analyze the penalties suffered and highlight recurring patterns in opposing shooters. In real competitions, the direct use of instant predictions is arduous due to regulatory constraints and tiny latency, but with sufficient calculation speed, a valid decision -making support could be obtained in the times previous to the stroke of the rigor. In practice, the IA could provide detailed information on the preparation of the shot to an assistant, who would turn the suggestions to the goalkeeper.
But be careful: the forecasts are obtained from historical data that do not always “frame” players well, who remain able to change strategies or to make unpredictable movements.
Without considering that when the players acquire awareness of the models, their techniques could evolve, developing even more effective anti-pressure patterns.
Limits and challenges. A similar study on the topic, conducted by a Cornell University team, has achieved even better results, proposing solutions that integrate the body and video sequences to then bet on the direction of the shot before the impact. The successful percentage was very high, up to 89%, but obtained in controlled tests, and not based on actual penalty pulled during a game. Finally, there are analytical sports platforms that already offer real -time forecasts, on the base, however purely statistical, that is, on the historian of the shots made by each individual footballer.
A penalty, however, remains a moment of strong unpredictability, in which a whole series of factors, including environmental factors, contribute to elevating the uncertainty. Furthermore, on the ethical level, questions are raised about the exploitation of technology in football: to what extent is it legitimate to use the IA to guess where the opponent will pull? In the game of the future, perhaps, ad hoc regulations could also be written to limit these innovations, and leave the healthy dose of unpredictability to the game that made it so popular.
