From Descriptive to Predictive: Unlocking the Potential of Football Data to go Beyond Tracking
25 years on from the first emergence of tracking data in football, computer vision technology is now unlocking its immense potential: merging it with live event data and in turn improving the way federations, leagues, professional teams and media analyse the game.
By Bradford Griffiths, VP Product Innovation, Stats Perform
On 25th February 1998, the French national team hosted a friendly in Marseille against Norway. The match was an entertaining 3-3 draw, but 25 years on is barely a footnote in the story of France’s first World Cup triumph.
However, in the context of football’s data revolution, this match is one of the most significant. This is because for the very first time, the on-field movement of every player on the pitch was digitally captured by an in-stadium tracking system. From this moment on, tactical analysis of football was never the same. Coaches and analysts now had tools which allowed teams to objectively scrutinise dynamic player movement, team shape and high intensity fitness.
In subsequent years, these tracking systems have evolved significantly: to the point where over three million data points are now being processed during every game, at a rate of 25 frames per second.
Going Beyond Tracking For Clubs
Despite advances in tracking systems, player tracking data has yet to be utilised to its full potential by professional clubs.
Whilst they have been able to use tracking data from their own league to profile teams, informing analysis, opposition scouting and domestic recruitment, they have not had ready access to deep tracking data to monitor teams or players beyond their own league.
But as we mark the 25th anniversary of that first ever match subject to in-game tracking, advances in AI technology means that tracking data is now on the verge of fundamentally changing the way the football industry utilises data, not just in terms of increased coverage, but also in how analysts navigate from descriptive data – using data to explain what happened during a game, to predictive data – predicting the most likely outcomes.
Going Beyond Tracking For Football Broadcasters
In the past decade, broadcasters have increasingly recognised the potential of leveraging data to create more engaging fan experiences for audiences who have grown up on a diet of fantasy games, player props betting, Football Manager and increased media exposure of stats and analysis.
Until now, most data-powered content, including advanced models such as Expected Goals and Live Win Probabilities, have been powered by ‘event data’, which in itself is a valuable and essential component to unlocking key stories. It comprises all on-ball ‘events’ that take place in a match, such as shots, passes and tackles. Since 2006 Opta has collected and timestamped the on-field location of every event in a game – which in some instances can exceed over 3,000 per match.
And until now, tracking data, in a broadcast context at least, has mainly been used to inform audiences about how far a player has run in a game, or the speed of a shot. This is in part due to challenges in unlocking real-time tactical insights from such a complex, detailed dataset, as well as challenges in synchronising player tracking with ‘event data,’ which provides the on-ball context.
This has meant that broadcasters have often opted to share insights, from both datasets, in isolation instead of using them together.
While deep, accurate ‘event data’ like Opta will always have the capability to deliver powerful insights on its own, utilising ‘event data’ in parallel with tracking data and at scale across global leagues will mean the data soon available to broadcasters will take on an entirely new level of sophistication.
We are now about see such an enormous step-change, thanks to a new AI-enriched dataset recently launched by Stats Perform: Opta Vision.
With Opta Vision, analysts can add even greater context to each event by knowing where players are located off-the-ball, and critically, their direction of travel and velocity.
Leveraging computer vision technology, Opta Vision comprises not only dynamic fitness data such as accelerations, sprints and maximum speeds, captured directly from remote video, but also a wide range of predictive AI models, trained on both remote tracking and event data, which can deliver previously unseen context of a game’s key moments. The applications for teams and broadcasters are many.
Off-Ball Performance & Richer Passing Insights
One of the new models available to broadcasters is Pressure Intensity. It uses dynamic off-ball insights, including the speed and direction of travel of defending players, to inform an audience of the impact of collective high pressure on the ball by a defensive team, captured all in real-time using tracking data.
For example, at the last men’s World Cup, five of the tournament’s top 10 players for committing high pressures came from surprise semi-finalists Morocco.
Without Opta Vision, storylines around their strong defensive organisation would rely on subjective observations. Now these insights can be automatically detected and shared with audiences as they are happening.
In addition, pundits can now use tracking-powered predictive models to deliver evidence-based analysis of not just a player’s performance but also their decision-making.
This includes decisions at key moments, such as a choice of pass – did they make the right decision when they had multiple options? Previously that would have been opinion-based and limited to perhaps one moment.
Now, with the availability of Pass Prediction outputs, pundits can evaluate not only the viable pass options, but also highlight which passes were most likely to be completed and crucially, increase the chances of a shot.
These types of insight are a logical but transformational evolution in how analysis can be utilised as a vehicle to inform audiences of what happened in a match, as well as adding a level of objectivity to discussions around what could have happened if different decisions had been taken.
Club Recruitment, Player Development & Analysis
In addition to enhancing broadcast experiences and increasing fan engagement, the ability to generate enriched data through computer vision means that scouting staff at clubs can now utilise dynamic tracking-derived insights to profile recruitment prospects in previously unavailable depth from leagues abroad.
From the start of 2023-24, Opta Vision data will be available from 40 leading competitions.
As well as scouting, there are other applications which have the potential to inform existing processes around post-match evaluation, opposition analysis and individual player development, including being able to better quantify a player’s on-ball decision-making.
25 years may have passed since that first successful test in Marseille, but with the support of AI and event data, today’s enriched tracking data is now ready to help the football industry make its next big jump forward in its ability to use technology to drive growth.
For more on Opta Vision, visit: statsperform.com/opta-vision