The Remote Tracking Manifesto

This is a biased text, written by a person who makes their living from selling tracking data to leagues and confederations around the world.

Even with its bias, I still think it could be an interesting piece for you, especially if you work for a football organisation that does not use any tracking data capture systems right now or that has tracking data collection deployed in their Tier 1 competitions/among their Tier 1 teams but not lower down the pyramid (lower divisions, youth, academies).

For a high-level view on the football data landscape and the direction it is heading, read my ECA speech.

Technology = New Types of Data Available

The times are exciting for football data & analytics. Things that once were done by humans with a piece of paper can now be done with hardware, and lately, only with software. More than that: we can now measure things that before were impossible to measure, simply due to the operational weight of these activities.

Tracking data is a great example. Tracking data is also called positional data because the main point of tracking systems is to capture the positions (i.e. coordinates) of all players & the ball at all times. How would you collect those coordinates without technology? Try to flirt with this problem for a while, and you will quickly see that it rapidly escalates from being difficult to execute and manage to being impossible*.

Positional data was indeed impossible to capture at the breadth and level of detail that is available to us today - thanks to tracking data collection technologies (also called ‘tracking systems’ or ‘tracking engines’).

The Barrier of Entry

The issue with the tracking systems has always been the high barrier of entry. Traditionally, to gather tracking data, football organisations would have to rely on hardware-based, multi-camera systems that would either be permanently installed in the stadiums or would be set up specifically for a given match (i.e. a portable tracking system). These tracking systems would not only be costly but would also result in a different set of pain points, depending on the type of football organisation that would use them.

Federation/National Team Perspective

For federations determined to track positional data for their national teams, the main pain points with the traditional systems were lack of flexibility and the operational hassle required to set them up. Federations could choose between two approaches to gather tracking data.

The first approach would be to rely on the tournament organiser, such as FIFA or UEFA, to provide the tracking system and the tracking data for each game within a given tournament.

This approach has proved to be challenging in its own way. Different tournament organisers would use different tracking data providers. Each of those providers would rely on different data formats, and that, in turn, would cause integration nightmares.

For one tournament you would receive data generated from the Tracab’s system; for another  tournament it would be the data from Second Spectrum’s system. Each tracking data and metadata file would be in their own, native format. The tracking data could be timestamped differently among various providers (i.e. which moment do we count as match 0:00:00?). To integrate this data into your own analytical infrastructure, whether a data platform, report format or spreadsheet reporting framework, would require additional work - often manual and often a lot of it.

That is if, as a federation, you get access to the raw tracking data in the first place. In the vast majority of cases you would be limited only to processed outputs (e.g. reports) generated by the tournament organiser. Essentially you would have access to only parts of the full picture even though somewhere out there, the full picture is available - just not to you.

The second approach would be to travel with your own portable camera system that would be capable of collecting the tracking data. This approach would give federations lots of independence and control, but it would also prove to be a logistical nightmare.

Mapping venues before each match, setting up and later calibrating the system - these are hours and hours spent just to collect the data for a single match and if things go wrong (e.g. an incorrect framing resulting in image quality loss or some sections of the pitch being out-of-frame; or situations as simple as cables not being connected properly and one camera angle not being recorded for the first 5 minutes of the match), all the work would be lost or a significant manual effort would be required to fix the footage.

League Perspective

For leagues the pain points would be different.

The pain point that I hear most commonly is the scale of the implementation. Deploying a league-wide tracking system is a complex project.

In the selection stage, whether through a Request for Proposal or through an independent search, the deployment requires careful consideration. The selected tracking system will be used not by one team of coaches, not by one team of performance analysts, but by all the teams within the league. The tracking system and its outputs need to meet the needs of all of the clubs, both the ones at the top of the table and the ones at the bottom of the table.

Commonly, the teams higher up the table will have the know-how and the data analysis/data science departments to work with raw data - whether tracking or any other type - while the teams lower down the table will expect at least some layers of post-processing applied to the datasets. The post-processing will make the tracking data more accessible and actionable for those that lack the capabilities to process the raw data themselves. The standard forms of post-processed outputs are reports and online platforms.

To further add to the challenge, a league-wide deployment is a significant project management effort. It ties together tens of stakeholders, both internal and external, and requires extensive cross-organisational cooperation.

Just to prepare and put forward a tender, the league needs to have a thorough understanding of clubs’ needs and readiness. The selection process then requires qualified people to review the Proposals and assess how well they fit into the league’s and clubs’ needs. Then finally, once the provider is selected, the implementation needs to happen: the tracking system needs to be deployed in all the club stadiums, which in most cases is an extensive project management effort.

Based on my conversations with leagues all around the world, usually it takes a few years from the moment that a league makes data collection and analytics part of their vision to the moment that tracking data becomes a tool in their or their clubs’ inventory.

The second pain point is the cost.

A league-wide tracking implementation means that each of the stadiums needs to be equipped with the tracking system that then needs to be maintained and operated. For countries that rely on only a handful of stadiums, this is less of an issue. For countries where each club has their own stadium, this becomes a significant limiting factor.

This pain point is further enlarged if you need to give access to the tracking system to youth or women’s clubs. These teams often play their matches in different stadiums than the men’s teams, which means that they will not have access to the tracking system and the data. (This is the case e.g. in the Netherlands, where many clubs’ youth & women's teams use the training grounds for the majority of matches).

The cost can also become prohibitive when the contract duration is taken into consideration. Possibly due to the complexity of the procurement and deployment processes, both leagues and providers tend to prefer multi-year contracts. While creating less overhead for both sides, the multi-year contracts also result in the league being bound to one provider, tying up a significant portion of the league’s budget for the following years.

Lastly, a league-wide deployment means months of full-time project management work. The management of this entire process is in itself costly. The cost is not only expressed through salaries paid, but there is also an opportunity cost, expressed as the time taken away from other, perhaps more pressing, club & league development initiatives that could be driven centrally by the league.

Often the leagues that would benefit from a tracking system the most, would not have the human resources to even manage a league-wide implementation project; not to mention then leveraging all of the benefits that tracking data can offer and teaching clubs how to reap them.

All of these disadvantages are accepted because tracking systems offer an exceptional quality of data.

Additionally, many leagues recognise the value of tracking data, and many of them are aware that these implementations will need to happen sooner or later if a league’s competitiveness is to be maintained or even improved.

Enter the Realm of Remote Tracking Systems

A partial solution to some of these pain points and common challenges is offered by remote tracking systems.

Remote tracking systems are optical tracking systems that, instead of relying on in-stadium cameras, use software to generate tracking data from the video footage that is readily available. These systems would commonly rely on either the tactical footage or, when the former is unavailable, the broadcast footage.

Remote tracking systems leverage the equipment that is already in use, most often by the video analysis teams employed by clubs and national teams, rather than asking leagues and federations to install new equipment. This way these systems can significantly reduce both the operational & capital costs and the deployment complexity.

They deliver the same end result, i.e. the dataset containing player & ball coordinates, but using a different method and a different set of tools. Rather than relying on multiple camera angles, these systems will use artificial intelligence (more specifically computer vision), geometry mapping (technical name: homography) and sophisticated modelling to redraw the pitch virtually and determine where the players are at every moment of a match. On top of that, a number of trained correction algorithms may be applied to a dataset to account for any mistakes. These algorithms would leverage machine learning, and, using a model trained over thousands of previously processed games, they would adjust the data accordingly.

Single-Camera Tracking ≠ Accurate Tracking Data?

A common objection associated with remote tracking relates to the quality of the data gathered by these systems. While this objection is sometimes partially true, it is important to acknowledge its dubious nature: the grounds for this objection can be found both in fact and in misconception. Let’s briefly analyse the two.

When objectively compared, the remote tracking systems can generate data as accurate as some of the multi-camera systems. This, however, varies among providers and heavily relies on the final use case for the data. For example, companies that specialise in scouting/recruitment data tend to compromise on the accuracy of their datasets for the sake of breadth of coverage. This is only logical, as coverage is highly desired for scouting - as a scout you would rather have low- to mid-quality numbers for ten leagues than have high-quality numbers for one league. This low- to mid-quality data, however, may not be the best for tactical, performance or player monitoring use cases. It may not be the best source of truth if you want to use players’ physical performance in the annual evaluation process. You really want to have as accurate data as possible in these scenarios.

The second nature of this objection lies in a common misconception. Remote tracking is often  associated and put on par with broadcast tracking. This mistaken association has serious implications for the quality of the tracking data and, if taken at face-value, will indeed mean that remote tracking generates lower-quality datasets than traditional, multi-camera systems.

Broadcast footage involves versatile camera angles, zoom-ins or even blind spots – the best and most familiar example being fragments of the match that are not broadcast due to a replay being played. Because of moments like that, broadcast footage can be tracked only to a limited extent, unlike tactical or panoramic footage where the majority of players are visible at all times and the camera angle is fixed and only moves horizontally.

In practice it means that out of the 90 minutes played, only, say, 75 minutes will be suitable for tracking. To add to that, within these 75 minutes we are unlikely to have the majority of players in-frame - which would be the case with tactical footage. This forces the tracking system to ‘guess’ the positions of the non-visible players. The guessing is done by trained machine learning algorithms - trained over thousands of match recordings where all players were visible at all times. It is, however, still guessing and does not represent an actual on-pitch situation.

It is important to emphasise, however, that remote tracking systems ARE NOT the same as broadcast tracking systems. Broadcast footage may be one of the usable video input types but is definitely not the only one. It is commonly used as a backup option - and serves that purpose well, as it is highly reliable in terms of the quality of production.

The most common footage type used for tracking is either wide-angle tactical footage or the Camera 1 footage - the former being common with any club/team that has their own video analysts and/or camera operators, and the latter being common with larger, international tournaments that often rely on an external footage provider to produce the match footage.

These footage types face challenges of their own - unrelated to the ‘guessing’ process - but simultaneously offer an almost full view of the pitch at all times, allowing tracking systems to generate accurate tracking data. This data will faithfully reflect the on-pitch situation.

Remote Tracking and Its Impact on the Football Ecosystem

This gets us to the finale of this manuscript: the meaning of the nuances of remote tracking systems and the role that these systems play in a wider football ecosystem - whether locally, on a national level, or globally, in international & intercontinental competitions.

Remote tracking systems are an important disruptor to the current status quo in football analytics. Their cost-efficiency allows access to accurate tracking data for organisations that could not afford such data before with traditional tracking systems. While they may not generate tracking data as accurate as a 32-camera tracking system, they can be deployed relatively quickly and at a significantly lower cost, allowing under-analysed leagues, competitions and matches to suddenly be tracked. They level the playing field by giving access to the same information to all participants within the ecosystem, including participants that could not afford to have this information in the past.

Furthermore, through working with some of the most technologically demanding football organisations in the world, we have found that for the vast majority of use cases, the accuracy of raw tracking data offered by a high-quality remote tracking systsystem willfully sufficient.

Organisations that manage to look past the misconceptions and validate the data themselves tend to appreciate the high quality of tracking data that can be obtained using the remote tracking systems. Through their own internal validation process, these organisations often find that the remote tracking data is comparable, and in some cases more accurate, than the data generated by a multi-camera system.

These organisations often leverage the cost-efficiency and the ease-of-deployment of the remote tracking systems to extend their tracking coverage across their local football pyramid. Such extensions would not have been possible before, simply because of the high costs and the logistical hurdles related to gathering tracking data with the traditional multi-camera systems.

If you would like to try a high-quality, high-precision tracking system, please reach out to me, the author of this article. At ReSpo.Vision we would be more than happy to run a demo for you and show what a FIFA EPTS-certified Remote Tracking Engine can do for you.