AI/ML at works behind the best FIFA2022
Last Sunday the world saw the best Worldcup football final (yes Soccer).
In this article, I would set the context on
What Football (Soccer) means to the world:?
For comparison, the 2022 Super Bowl pulled in nearly 100 million viewers, whereas 3.572 billion viewers which is more than half of the global population tuned in to world football’s ultimate competition. Soccer is the world's most popular sport, played by over 250 million people in more than 200 countries. Fifa 22 tournament was the last with 32 participating teams, with the number of teams being increased to 48 for 2026?
The Lusail Iconic Stadium in Doha plays host to the 22nd FIFA World Cup Final as two nations attempt to inscribe themselves into football history.
For first-timers or regulars, the occasion is just as grand; just as significant.
Football's ultimate game, however, has often disappointed over the years. With the two finalists so close to reaching the pinnacle, the primary focus is typically avoiding defeat as opposed to winning the game. But when Lionel Messi and Argentina took on Kylian Mbappe and France in the final of the 2022 World Cup it felt like the peak Roger Federer vs Rafael Nadal in the 2008 Wimbledon final. It felt like Lewis Hamilton vs Max Verstappen in the final race of the 2021 Formula 1 season and it felt like Ben Stokes and England vs Australia in the 2019 Ashes. Unbelievable theatre from the absolute best in the business.
Gonzalo Montiel sealed the deal with Argentina's fourth penalty, ending the greatest World Cup final ever played.
Argentina 3-3 France (Argentina win 4-2 on penalties) - Can't get better than this.
Challenge in Decision Making for match referee:?Since Football is a low-scoring game, any error can change the game outcome devastating the team, players' careers, coach's careers, or country standing, and even worse can anger fans in and out of the stadium, leading to Riots. Secondly, the offside rule mandates that during a move, an attacking player, when in the opposition half, must have at least two opposition players, including the goalkeeper, between him and the opposition goal when a pass is being played to him.?There are 8 types of such possibilities.?In a nutshell position of the ball and players for both sides contribute to offside and hence prone to error with an on-field referee or in-line cameras. FIFA resorted to AI/ML to bring transparency to decision making
How AI/ML helped in Decision Making @FIFA22:?This world cup not only saw the best Football match but also the best AI/ML technology at work. Accurate decision-making requires the exact position of Football and Players at all times in 90 minutes games (120+ mins including extra time), the amount of data collection and processing is enormous. Infrastructure@scale can be explained in the following 4 components
FOOTBALL DATA ECOSYSTEM?- The FIFA Football Data Ecosystem is a complex network of several data sources, data processors, and distribution layers providing consistent and high-quality data to all relevant stakeholders.
Collection of positional data- To collect the positional data (x-y coordinates) of all players, the referees, and the ball, a state-of-the-art optical tracking system will be installed in all eight stadiums. The optical tracking system is able to capture player positioning multiple times per second, accurate to the nearest centimeter. This data not only reflects player position but can also be used to calculate speed, distance, and direction of play.
SEMI-AUTOMATED OFFSIDE TECHNOLOGY -?One of the main challenges in the development of advanced offside technology is the accurate and automated detection of the kick-point. Possible solutions — notably, tracking data from sensor technology or video data from camera systems — were considered.
?the new technology mounts 12 dedicated cameras beneath the stadium roof to track the ball (which also has sensors) and up to 29 data points for each player, 50 times per second, to calculate their exact position on the pitch. The 29 collected data points include all limbs and extremities relevant for making offside calls.
Following are constant data rates sent from the field, processed and trend/result sent back to the on-field referee (for data center infrastructure required refer to NALSD blog in Reference)
Players Data Rate 29 x 1KB x 22 x 50 = 31,900 KB/sec ~ 32MBps ~ 256Mbps
Data Storage for 120 mins match = 230 GB - [with a replication factor of 3] ~700GB
Furthermore, a system has to correctly identify which body part places a player onside or offside. Accuracy tests have shown that human operators tend to pick different body parts for offside lines. Strides have been made in that area as well, with the automated system providing learning to correctly model a player’s skeleton. In the future, the developed algorithms of the system should be able to automatically identify which body part placed the player offside and by what distance.
The official match ball will provide a further vital element for the detection of tight offside incidents as an inertial measurement unit (IMU) sensor will be placed inside the ball. This sensor, positioned in the center of the ball,?sends ball data to the video-operation room 500 times per second,?enabling very precise detection of the kick point.
Ball Data Rate = 1KB x 500 = 500KBps ~ 4Mbps
领英推荐
for 120 mins[7200 sec] match = 4Mbps x 7200 = 28,800 Mbps ~30Gb
By combining the limb- and ball-tracking data and applying artificial intelligence, the new technology provides an automated offside alert to the video match officials inside the video-operation room whenever the ball is received by an attacker who was in an offside position at the moment the ball was played by a teammate. Before informing the on-field referee, the video match officials validate the proposed decision by manually checking the automatically selected kick point and the automatically created offside line, which is based on the calculated positions of the players’ limbs. This process happens within a few seconds and means that offside decisions can be made faster and more accurately.
GOAL-LINE TECHNOLOGY-?The system uses 14 high-speed cameras mounted on the catwalk of the stadium/under the roof. The data from the cameras is used to create a 3D animation to visualize the decision to the fans on TV and on the giant screen inside the stadium.
The average time for a VAR offside check currently is about 70 seconds. Semi-automated offside technology gets this down to between 15 and 25 seconds in FIFA’s tests, with further improvements technology can get an answer within four or five seconds.
Within seconds 3D animation needs to be generated, and limb position manually checked by Video match officials communicated to the on-field referee upon confirmation displaying the same 3D animation to broadcast and spectators. Compute requirement is enormous
Finally, businesses have already started minimum digitization for achieving a?higher-level of AI-driven requirements?for the World Cup 2022. Without proper data, machine learning platforms won’t be able to produce outcomes of value. Medium to large organizations started applying Artificial Intelligence and Machine Learning concepts that their data needs to be in a usable state. Once digitization, say as a stepping stone layer is initiated, cognitive technologies are added to extract useful value and gain intelligence on software dashboards or other third-party integrations to have efficient results.
Return of Investment with respect to Viewer Experience-?I believe in Landing vs Launching, which means what was the impact of technology on the game. I would let numbers speak for themselves here
Total overturns: 25
Rejected overturns: 2
Leading to goals: 6
Leading to disallowed goals: 10
Penalties awarded: 10 (6 missed)
~ for holding: 2
~ for handball: 2
Penalties canceled: 1 (offside)
Penalty retakes: 1 (GK encroaching)
Rejected penalties: 2
Goals ruled out for offside: 8
Goals after incorrect offside: 2
Red cards: 1 (out of a total 5, guided by AI/ML Algorithm)
Reference:
2. https://www.espn.com/soccer/fifa-world-cup/story/4807433/var-review-every-decision-at-the-world-cup-analysed