Sport is one of the sectors in which Artificial Intelligence is seeing widespread innovation . The Tokyo 2020 Olympic Games (unfortunately postponed until the end of July 2021), were to be the first event to make large use of Virtual, Augmented and Mixed Reality. The Olympic Games committee, toegether with Intel, was supposed to provide services for spectators such as 3D athlete tracking at the sporting events. An AI facial recognition software called Neoface was to be used in Tokyo to identify 300,000 people, including athletes, volunteers, journalists, camera workers, and security personnel, who were involved in the games.
Let’s hope that at least the Japanese fans will be able to enjoy this possibility since, for the moment at least, foreign fans aren’t expected to be able to attend, due to the pandemic.
In the following paragraphs, I’m going to tell you what the most interesting applications of artificial intelligence are, and also adding what I think is the real technological challenge.
In Formula 1, continuous technological improvement contributes to increasing not only vehicle performance, but also and above all the drivers’ safety. Amazon AWS is the official provider of both cloud services and machine learning technology. During a Grand Prix, each car carries 120 sensors that generate 1.1 million telemetrc data points per second, which are transmitted from the car to the pits. This data together with more than 69 years of historical racing data stored in S3, the cloud storage service on AWS, to make it possible to explain the unique decision-making process conducted in the team boxes to both fans and teams: how many laps will Hamilton need to get to Bottas? Which laps are best for the Ferrari car to change its tires? Is it possible to analyze the results of Friday’s free practice to predict what will happen in the official Saturday trials? The original article explaining the model used by F1 Insight is available at this address.
Player Performance and Augmented Coaching
Artificial intelligence is increasingly becoming a tool for improving athletes’ performance. Homecourt is a personal trainer for basketball players that makes use of Computer Vision and Machine Learning which, through the device’s camera, is able to capture a player’s ever movement and shot and analyze them in real time, then providing feedback: the percentage of shots made, speed, angle of release, elevation ,and ball handling.
IBM, in partnership with Red Hat and the United States Tennis Association, launched a product in 2019 called Coach Advisor for junior player coaches and talent scouts. Traditionally, coaches relied on their experience and instinct to assess an athlete’s mechanics and endurance. However, a match can last several hours with movements covering up to 10 km along the entire field: quantifying the effort and energy expended is not easy. So special datasets are used:
- Physiological load: measure of the player’s physical effort and the overall work done in a match
- Mechanical intensity: measure of accelerations and decelerations throughout the game.
Since good physical health is of the utmost importance during an athelete’s career, for both amateurs and professionals, in recent years there has been an explosion of the so-called “wearable devices” that collect the main performance parameters and store them on a Cloud. What can be done with this data? Preventing muscle injuries or discovering more serious hidden problems are just some examples of the possibilites. In reality, we are still at the beginning of the reality of consolidated and possibly democratic algorithms: that is, those affordable for ordinary people not only for millionaire teams and organizations.
Streaming and broadcasting of events
An AI-based system can automatically choose the right camera angle to show on users’ screens, or provide subtitles in different languages based on the viewers’ location. Deep Learning now makes it possible to create fully automated television productions that are at the same level as professional filming. And these shots can be used to extract highlights that are later distributed to channels that do not have exclusive rights to the event. Science fiction? IBM Watson has been helping the Wimbledon tennis tournament to extract the best moments of its matches for years. For the 2019 edition of the tournament, IBM said it instructed Watson to use acoustic data to measure the impact of the ball and achieve better cropping of the image. A company like Pixellot offers specialized hardware coupled with AI software to record sporting events and extract their most significant moments. The final step is to automatically extract a written report from these video summaries: it is the so-called automated journalism that uses Natural Language Processing. An example of such a platform is Wordsmith.
Are you curious to take a look at some code and not just commercial products? Deltatre is one of the technology leaders providing services for major sporting events and federations. Its research unit has created a github account which hosts several projects including Action recognition in sports (soccer, shots) started in collaboration with Microsoft. It is a solution for the automatic detection of football actions (for example when a player shoots on goal) that combines numerous computer vision and deep learning techniques, and is performed in a cloud-scale pipeline, using Microsoft Azure and open-source frameworks such as Pytorch, TensorFlow and OpenCV. You can follow their YouTube channel for interesting insights.
By now we’re using to seeing the use of goal line technology or VAR during football matches as it’s been around for several years. However, these are quite trivial applications. There are, on the other hand, more interesting cases such as sports in which it is necessary to assign a judgment based on the difficulty levels of an exercise: think for example of rhythmic and artistic gymnastics, or diving. Fujitsu, on during their world championships, presented a device the size of a Wi-Fi router containing three-dimensional laser sensors to track the gymnasts’ movements. The data, processed by AI models, offers judges and referees speed, angles, and body positions; helping them to interpret the exercises correctly.
Strategic analysis: the real challenge
Marcelo Bielsa, a well-known football manager currently at Leeds United, talked about how he prepares his team tactically. His analysis team watches all of the opposing team’s 51 games in the current and previous season. Each game requires a four-hour analysis in search of very specific information: starting players, reserves, tactics, and strategic decisions. A methodology that takes about 200 hours for each game leading to subjective and often inaccurate conclusions. Can technology make this process more efficient? Well, it’s not an easy task. Suppose we collect all data relating to individual players. Unfortunately, the strategies and lineups are hidden by the tracking data in which the players, obviously, are moving around: it is therefore a classic case in which you have to separate the background noise from the data. All you need is enough computing power to be able to query tracking data on a very large number of theoretical combinations.
The sports field is showing us some of the most interesting advanced applications for artificial intelligence, both those aimed at improving the athlete’s performance and those designed for improving the spectators’ experience in a stadium or at home in front of the TV.
We just have to wait for the Tokyo Olympics to see what will be on display, noting that all these examples show us what the purpose of these technologies should be: to help us to improve our life, in all its aspects.
Keep following us!