As we saw in our previous articles in this series, there area great many different data sources in sports. Making good use of that data to create more compelling and engaging TV is down to the art of graphic visualization. Using images instead of words to communicate complex information goes right back into the depths of human history – the Lascaux cave paintings are over 17,000 years old and some say they depict the constellations of the stars in the night sky using animal graphics.

The practicalities of creating graphics for live sports is a game of three halves; it is part mathematics, part visual design, and part production workflow… but these are all driven by a need to tell a story. The story itself can be many things; discussion & history of league standings, analysis of previous historical competitions, conveying the in game/race/match statistics to keep the viewer on top of what is happening, or the story can be how to help an expert present pre/post-match/race/game tactics and performance analysis.

As we will see in subsequent articles where we talk to designers, these graphics are typically custom built for each sport each season and usually by a specialist provider or in house team in response to a client or key stakeholder brief. The software and methods used vary by sport and provider and how these integrate with live production will also be explored later in the series. Here we take a look at the history, various types of visualization and the central role of AI/ML in this key niche area of production technology.

Match, Race, League, Team & Player Statistics

The Sumerians were using tables to present data in rows and columns on clay tablets 4400 years ago. It is pretty difficult to put a date on when they were first used for sports broadcasting, but it was long enough ago not to matter – official league or tournament tables have been a staple of sports TV pre and post event coverage for many decades. The bar and line chart were invented by William Playfair in 1786 and they too are a permanent fixture of statistical analysis of individual competitors or clubs and teams.

The early days of cards and flip charts have given way to very slick animations of tables and charts that update dynamically in real time to reflect results or current league or tournament standings. Infographics can be 2D, or increasingly 3D and can be presented as classic 2D visuals or as Augmented Reality overlays. They need to have an operator interface to control the presentation live, and in some scenarios, provide capacity for talent control using touch screens or tablets (digital, not clay). In the expanding live multi-platform delivery landscape, graphics of all kinds also need to be responsive and adapt to the size and orientation of the viewing device.

Responsive design is not to be trivialized when it comes to infographics – If your team is cutting different packages for different delivery channels, infographics may need to be re-configured and re-applied during the edit workflow, and that must be taken into consideration early in the development of designs & live workflow.

Pre and post-game presentation graphics are data driven but not always drawing on live/real time data feeds. They obviously can be API based live feeds pulling from third party sources like leagues or governing bodies, or specialist sports analytics companies, but because they are usually part of a presentation not a real time screen overlay, they can also be a visualization that pulls data from a database, CSV file or other local or watch folder data source.

The mechanics of production are relatively simple in the sense that they are well known and relatively standardized (give or take the nuances of the various vendor operating environments); ahead of the live production the design team create a graphic visualization (or adapts a template) that pulls part of its content dynamically from a data source, and the data updates alpha-numeric fields or adapts and/or animates graphs. All of which is usually done within the live production software environment to deliver feeds to the switcher. It can however also be done using a stand-alone specialist data driven graphics package or created from scratch by coders working in python or other application development languages.

Part of the great skill of the best sport experts is the way in which they draw in historical knowledge to enhance their commentary. There are a number of specialist software systems available that combine the discipline of graphic design with statistical analysis and many sports use this to great effect.

The example in Figure 1 is a two-dimensional graphic created in Tableau, which we are using as an example because it is a well-known, template based, dynamic data visualization design tool, but it is not really one of the main sports graphics vendors and we are avoiding endorsing any specific provider. The graphic is actually an interactive HTML 5 deployment, based on rich data drawn from soccer analytics specialists WhoScored.com to deliver a well-presented and very desnse set of statistics for a single game.

It works because the data was gathered, mathematical statistical analysis was performed on the data, the results were input into a dynamic graphics application, which somebody had coded to transform the raw data into visuals… but most of all it works because somebody had a vision of a good story and the designer did a good job of creating a visualization to support it. Figure 2 is equally good - this time historical data reveals visually how changes to the sport itself changed competitive outcomes in the Tour De France.

Live Data Driven

As discussed in article 1 of this series, pulling live data feeds from official race or match systems has been happening for decades – like the Formula 1 live race standings table in their side-bar – and then F1 over the years adding rich data feeds from the car systems themselves to the tables to create a more immersive fan experience. Many sports have also been drawing in ML powered statistical analysis to deepen fan engagement – a classic example being Wimbledon and the introduction of ML powered real time tennis statistics to coverage back in 2012. The early introduction of ML to do the mathematical heavy lifting of statistical analysis is a pivotal point in this evolutionary tale.

A great sports expert presenter combines their knowledge with their own emotional connection to the history of the sport and that is unique and to be cherished – the style and empathetic connection of the audience to the presenter is the stuff of legend.

The relentless rise of AI brings something very different. Given proper training, ML can bring something close to the entire statistical history of a sport to bear on the analysis of the data, to create deep insight into player or club performance etc, and it can do it in real time as the event is happening. The combination of the human expertise and the machine insight is a powerful tool in the creation of compelling production.

Play Analysis

Data enhanced play analysis is all about using icons on a screen to depict the positions of the stars on the field or track.

Simple 2D layouts for formations can be created and animated in a number of ways and are relatively simple compared to animated overlay graphics. What started out as using a stylus to draw rudimentary rings around athletes and arrows to show which way they should have run has evolved significantly in recent years. The capacity for ML to rapidly and reliably identify ball, players, cars etc and relatively precisely track their movement has changed everything.

There is a direct workflow connection between camera array tracking at events and play analysis in the sense that the camera array is often used to synchronize multiple camera angles for instant review in replay systems, and the replay systems feed clips to the live graphics production software.

As discussed in article 1 in this series, different vendor applications take different approaches to the creation of object tracking based play analysis visualizations. Some take the data feed from the array system to use as object tracking data and combine this with clips from replay to create visualizations.

Others do not draw object tracking from array systems, they do their own ML powered image analysis. Video clips are drawn from replay. The camera angles for a specific production need to be calibrated ahead of the live production. ML helps identify players or objects and generates object tracking data.

Regardless of how the object tracking data is derived the graphics workflow is broadly similar; a template library of visual sequences is used to streamline the placement of graphic overlays onto clips. Templating enables a significant selection of different types of data visualization, along with multiple virtual camera angles to support it. The complex mathematical analysis required to make such play analysis visualizations work is not trivial and not a DIY exercise, so this is one area where working with templates created by a vendor as part of their application is fairly inevitable. What you get varies in type and style by vendor. It all has to be done rapidly during a live production to prepare sequences in response to requests from the ‘talent’ (expert guests) in time for half time and final whistle presentation production. We will discuss how they are generated and controlled during live production later in the series.

Most mainstream sports are now supported by a range of templates found in various vendor applications and the breadth and depth of the effects available has genuinely transformed coverage. For the viewer it feels like a convergence of gaming and viewing, which in many ways it is. In the fiercely competitive broadcast landscape of live sports, raising the bar of the immersive experience in analysis is one of the more competitive areas of production. It has also become big business – the arrival of ML powered virtual environments has created new and ingenious monetization opportunities where branding can be placed dynamically amongst the action and adapted for localization.

ML Calling The Shots

What Machine Learning does really well is statistical analysis. ML is what is calculating and displaying the acceleration of ball, player, car etc in real time. ML is creating smart visualizations of all the positions and angles from which a given player is scoring goals… etc etc. Generative AI is already able to calculate the statistical probability of passes and plays completing successfully. Based on historical statistical data, it can already predict the percentage probability of what the next play may be given the current position of players and ball. It seems inevitable that vendors will continue to find ways to leverage this capacity for statistical prediction to automate more of the entire sports production process.