Current AI/Machine Learning trends in Football

8 min readSep 16, 2020

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Football is considered more than a sport for many fans all around the world. It keeps one’s engagement at check all the time, by providing a rollercoaster of emotions. There are numerous leagues out there in the world, consisting of various teams with diverse backgrounds, which unites us all through the spirit of the game. With the advancement in technology, in this beautiful game, there lies countless possibilities to enhance the viewing experience for fans, playing experience for players, managing and administrating experience for coaches and staff.

Reminisce the smell of the field in your last playtime and the adrenaline boosts during the game !!!!!

Artificial Intelligence (AI) is a methodology in which we train machines to mimic human behavior by simulative actions. On another hand, Machine Learning is a subset of AI, which doesn’t require explicit programming but is used to learn through experience. Further, the core element-binding AI, by aiding research and analysis for automation in tasks is ‘Data’. We are progressing and evolving steadily in this ‘Information age’ on towards a more ambiguous state, which will be in constant flux because of the rapid development and penetration of technology into various fields. Subsequently, I believe that there is more boon than bane in the advent of these new technologies in this world and it will eventually rejuvenate our mere nature of humanity.

“If we do it right, we might be able to evolve a form of work that taps into our uniquely human capabilities and restores our humanity. The ultimate paradox is that this technology may become a powerful catalyst that we need to reclaim our humanity.” — John Hagel

My goal here is to visualize on the optimistic side of AI/Machine learning and its implementation in the field of Football. Multiple-use cases will be discussed along with its potential applications in future trends.

Where and How AI/Machine Learning is being used in the current trend?

Generic Stereotype of an image of machine learning applied [1]

Kickoff.ai

They use machine learning to predict outcomes of matches based on past data collected by using a bayesian interface. The outcomes are predicted as the conditional probability of past events to evaluate a team’s winning rate [2]. One such example is shown in the figure given below:

Match outcome analysis based on dynamic data input of each team

The picture was taken from their website, in which the winning probability of teams is predicted by their bayesian model, by taking in dynamic inputs of data from the team over time.

2. Loughborough University (Computer Scientists)

Secondly, This project aimed to replace the current performance analysis used on players, which is an extremely time-consuming process, with limitless video-readings of players and assessing their performance to gain intel on events in-game. The pre-existed methods are also deemed to be more labor-intensive, inconsistent, and tend to be more biased on perceptive cognition of the human’s handling the data. To overcome these issues, the lead director along with their team developed a hybrid system of using camera-based automated methods, coupled with computer vision/deep-learning to aid low-cost to time performance measures in data analysis. Their major goals were[3]:

Detecting body pose and limbs

This involved using deep convolutional layers to learn hidden patterns and extract distinct features from a large amount of data such as thousands of match recordings, consisting of multiple teams, camera poses, gait movements (inclusive of running, walking, kicking).

Tracking of individual gait movements (pose of players) [3]

Tracking players to access individual performance stats

Accessing and geo-locating each player on the pitch with relation to the other gives scope to enhance coordination between one another in this team sport.

Camera stitching

Limited field of view of the cameras for analyzing college games was an issue and hence they devised a low-cost robust GoPro based automated camera stitching methodology with the help of their industry partner ‘Statmetrix’, which is now made commercially available for performance analysis.

Tracking of players by the automated camera-stitched FOV [3]

3. SciSports

Another company which looked promising is SciSports and their machine learning capabilities give management and professional teams to track player’s performances and aids in scouting them on a whole different level. Their data-intelligence software keeps track of more than half a million players all around the world, searching earnestly for young budding talents to keep track of and report to their clients, aiding them for recruitment analysis. More importantly, it is the vision of the creators that guides the company in reaching its full potential, in an interview with Giels Brouwer (Founder & Chief Innovation Officer of SciSports) [4], he speaks of how he was fascinated by playing ‘Football manager’ and wanted it to transcend from software and make it a real-world application. He further feels that the data scouting platform that he had created complements well with the on-site scouting held by many teams, as it needs one and other to save time and money. Let’s have a look at the UI of their software [5],

This shows the home-screen of their UI [5]

The above pictures not only show each player’s stats and playing style, but it also forecasts their potential and what quality they can improve on, to better benefit their style or for the team that they are going to be recruited for.

4. GAMEFACE.AI

They are a company that provides an instant gathering of data points such as goals, free kicks, fouls, and shot on goals from frame to frame of the video uploaded for analytics.

Important key events of in-game are gathered by the software [6]

Further, the exact coordinates of each of the players are spatially located according to their coordinates in real-time, with an additional interface for their corresponding heat-maps for analyzing the positional intel of players.

Positional intelligence provided by geo-tracking and heatmaps [6]

Additionally, one other awesome feature that aids the management is ‘searching for custom-created traits’ for each associated players, for example

You can create and search for key events in the game [6]

5.SPORT LOGIQ

Additionally SPORT LOGIQ redefined all the above-stated applications with a fantastic UI-based database for clinets. They provide physical and contextual logic details from the broadcasted game such as player positioning and translation of players in an x-y coordinate frame, game’s camera FOV tracking inclusive of the ball being tracked real-time. They target the global scouting market, player recruitment, and analysis for teams [7].

One such example of their contextual team analysis software interface is shown below:

This UI provides a database of the in fed video streams, which with the help of CV and deep learning aids in segmenting and allocating it in a database format for analysis. Further, a segment of phase-of-play (eg: Ball possession (BP)-> Ball possession opponent(BPO)) which happened in the game is represented in a matrix format, which can be then selected to view the trimmed video which shows the video for analysis in the output window. Subsequently, in-game events are categorized and displayed in a comprehendible manner by the UI for researching breakthrough passes, pressure building situations, a passing network of players, etc.

In-depth strategic feedback for team dynamics [7]

Additionally, they provide player analysis from anywhere in the world by adopting machine learning algorithms for studying event-based player tracking by relying on their physical metrics. This can aid in the recruitment of players to any club in any league, not limited to just relying on the player stats but also his/her adaptability in the corresponding league. For example, one such case-study (Timo Werner’s arrival to Chelsea Football Club) of theirs shows:

Comparison of the player’s stats with an existing player and past data from both leagues [7]

This example shows the player’s stats being compared to the arriving club’s existing strikers, former league’s strikers, the incoming league’s strikers. This provides an insight into whether the incoming player will be able to adapt to the league and in maintaining his stats.

6.Sports Technology Lab (with Preferred Networks Inc)

Hakuhodo DY Holdings Inc. and Hakuhodo DY Media Partners Inc. merged to form ‘Sports Technology lab’ which aids in providing innovative deep-learning-based analysis in collaboration with ‘Preferred Networks Inc.’. They developed the ‘PitchBrain’ which is a specialized football-based analysis tool providing three primary functions:[8]

automatic scene detection and tagging: Novel computer vision algorithms are used for body pose estimation
Team-style categorization
passing options visualization

PitchBrain in action and component analysis of their 3-feature algorithms [9]

Tracking chips and sensors help in determining the pose-estimation of players, which aids their model in training, by taking in inputs of players and ball positions. This helps in gaining semantic-level predictions of scenes according to the movement of the player gaits.

7. Other Deep learning analysis performed on football

When researching on potential usage of deep learning in the field of football, I came across this wonderful footage of Convolutional Neural Networks (CNNs) and Long short term memory (LSTM) used in conjuncture for training AI to play FIFA by ‘Chintan Trivedi’.

This video shows FIFA being played by an AI [10]

There is also his blog page in ‘towards data science’ (https://towardsdatascience.com/building-a-deep-neural-network-to-play-fifa-18-dce54d45e675) which shows the entire working principle of this system, which on a higher level is shown as in this picture from his blog.

Generic system flow of deep-learning based FIFA [11]

The potential for such a project to grow is enormous and enables multi-level prototyping possible with such systems to mimic strategic events in a simulated fashion for post-match analysis.

Summing up

Surge in AI-based techniques such as Machine learning and deep learning acts as a huge beneficiary in sports analytics, aiding countless services in any particular domain. Especially in the field of football, we can see countless companies using it to garner revenue by providing in-game analysis for football teams, which aids their management staff, scouting network, recruitment team, head coach, and their financial bodies. Growth is natural in this field and I can see with the help of technology, huge potential for reaching un-foreseen heights in this beautiful game of FOOTBALL.