Tracking data — the most detailed and accurate information about players actions on the pitch
Tracking data, which comprises the spatial coordinates of all players and the ball at every moment, serves as a foundation for constructing highly accurate models that estimate a range of football metrics.
In contemporary analysis, this data is instrumental in dissecting team tactics, observing changes in formations throughout the game, and closely monitoring players’ movements during set pieces and off-ball actions, among other aspects.
1. Standard Video-Based Tracking Techniques
Conventional video-based tracking, a complex process, requires gathering detailed data. The prevalent method employs an optical tracking system with a camera array. The configuration and number of cameras may differ, yet all systems rely on a stereo pair as their core. In its most basic form, it uses two cameras, but more sophisticated arrangements include two sets of cameras. These setups capture a stereo image of both players and the ball, allowing for the precise calculation of their coordinates through triangulation.
A leading solution in this field is the TRACAB optical tracking system from ChyronHego, known for its broadcasting software. TRACAB is divided into two main product lines:
- Gen4, the second-latest generation, which includes a single stereo pair made up of two groups of three cameras.
- Gen5, the latest generation, with two stereo pairs opposite each touchline and two conventional monocular cameras positioned behind the goals.
TRACAB Gen 4 has been deployed in more than 200 stadiums globally and was utilized in the EPL from 2013 until 2020 when it switched to Second Spectrum, as well as in the Bundesliga, La Liga, and other prominent championships. The introduction of Gen 5 in 2019 marked a significant advancement, with it now being actively employed in the Bundesliga. As of the middle of 2020, 36 stadiums in Germany had adopted this advanced optical system.
In an effort to gauge the effectiveness of the TRACAB system, a comprehensive series of tests was carried out at the Esprit Arena in Düsseldorf two years ago. These tests employed specialized equipment, including infrared cameras and tags, along with corresponding software provided by VICON. An exclusive area on one half of the pitch was designated for this experiment, encircled by 33 infrared cameras that served both as emitters and detectors of infrared radiation.
Over a span of two days, 20 players from the Oberliga, the fifth level of the German football league hierarchy, engaged in a series of football drills. Each player was outfitted with special infrared tags designed to reflect the infrared radiation emitted by cameras.
Prior to the initiation of the experiments, a meticulous camera calibration was performed, leveraging precise data on the distances between various pitch markings such as the penalty area and center circle. This calibration process culminated in a final measurement error margin of approximately ± 2 mm.
Specialized software processed the images, capturing the 3D coordinates of all players with a remarkable frequency of 100 Hz (or 100 measurements per second), using these measurements as benchmark references.
At the same time, the TRACAB system was operating in the stadium. It also measured players’ coordinates but at a lower frequency — 25 Hz. A detailed description of the comparison between TRACAB measurements and reference values can be found here. We are more interested in the final results:
- The TRACAB Gen4 system exhibited an overall coordinate measurement error of 9 cm, whereas Gen5 showed a slight improvement and achieved 8 cm. Errors for various movement types, including slow movements, direct line acceleration, and abrupt directional changes, were assessed individually.
- The accuracy of speed measurements showed an overall error of 0.09 m/s for Gen4 and a slightly better 0.08 m/s for Gen5. Considering that the peak speeds of the fastest footballers can hit 34–36 km/h (9.5–10 m/s), this error rate represents approximately 1% of their maximum speed.
A critical insight from these findings is the 8–9 cm error margin in coordinate measurements achieved by TRACAB systems, setting a significant benchmark. This performance starkly contrasts with results from analogous experiments conducted years prior on the SportUV system by Stats Perform, where VICON technology also provided reference measurements. The SportUV system exhibited a substantially higher error margin of approximately 56 cm, making it seven times less accurate than TRACAB.
2. Beyond Optical Tracking: Alternative Systems
There exist two notable alternative methodologies for gathering tracking data.
2.1. LPS (Local Positioning System)
This system involves the strategic placement of specialized equipment, termed base stations, around the perimeter of the stadium. These stations emit radio signals that bounce back from both players and the ball. To facilitate this process, each player is equipped with a transponder, a device designed to reflect the signals emitted by the base stations, with a similar device embedded within the ball itself.
The core mechanism of LPS hinges on the precise timing of signal transmission and reception by each base station. Given that the exact locations of these base stations are known and fixed, the system is able to accurately calculate the real-time coordinates of every player and the ball, thereby generating comprehensive tracking data.
Typically, LPS solutions exhibit an error margin ranging between 19 and 27 cm. For detailed insights into these figures, reference links are available here. Among the examples of LPS technology in action is INMOTIO, renowned for its sophisticated equipment.
2.2. GPS (Global Positioning System)
Global Positioning Systems operate on the principle of attaching GPS trackers to players and incorporating them into the ball. These trackers engage with signals from a constellation of satellites, employing the satellite coordinates and signal transmission time to deduce the precise locations of players and the ball. Despite their wide application, GPS tracking systems suffer from a significant drawback: their accuracy, which can err by as much as 1 meter.
Given these limitations, optical tracking systems are currently favored for their superior accuracy compared to LPS/GPS alternatives. Camera-based systems also have the advantage of not requiring players to wear additional devices, a significant benefit. Nevertheless, access to such advanced tracking technology remains a challenge for many, especially among lower-tier football leagues.
3. Considering Alternatives for Financially Modest Clubs (Beyond the Elite Leagues)
Machine Vision-Based Tracking data
The landscape is gradually being reshaped by emerging firms that propose accessible tracking solutions, sidestepping the complexities and costs associated with multi-camera optical systems. These novel approaches, which range from using a singular point of capture to dispensing with external hardware altogether, promise to democratize access to sophisticated tracking data. Below, we highlight a few of these groundbreaking solutions.
3.1. SkillCorner
Hailing from France, SkillCorner has introduced a method of gleaning tracking data directly from broadcast images in real-time. This process, powered by specialized machine learning algorithms for video analysis, successfully extracts 2D coordinates for players and 3D coordinates for the ball.
A notable limitation of this approach, however, is its inability to account for off-screen action, as illustrated by the comparative analysis of SkillCorner and traditional optical tracking data. While SkillCorner’s metrics align closely with those of optical systems for the majority of a player’s movement, tracking ceases when the player moves out of the camera’s view.
In an effort to bridge this gap, SkillCorner’s engineers are experimenting with models to estimate player positions in parts of the field not captured by the camera, especially those not directly involved in the ongoing play.
For a more in-depth look at SkillCorner’s innovative approach to data collection, the Friends of Tracking YouTube channel offers a detailed presentation. Introductory data from SkillCorner is also freely available for those interested in exploring this technology.
SkillCorner’s data transcends its utility beyond the realm of financially constrained smaller clubs, extending its advantages to football giants. Liverpool, for example, while equipped with optical tracking data from the official EPL data provider, faces a scarcity of similar data from other leagues. In an age where comprehensive tracking information is vital for scouting, SkillCorner emerges as a strategic asset.
This is exemplified by Liverpool’s strategic acquisitions of Tsimikas, Alcantara, and Jota during the 2021–2022 EPL season, with SkillCorner data playing a pivotal role, particularly because Liverpool previously only had access to Premier League-specific tracking data for Jota. Engaging SkillCorner allowed Liverpool to overcome the shortfall of international league data.
3.2. Metrica sports
Dutch-based Metrica Sports stands at the forefront of software solutions for the football industry, pioneering the development of ML algorithms that adeptly identify tracking data from a myriad of sources, including TV broadcasts and drone imagery.
Their website claims a data accuracy error of approximately 10 cm, a figure that seems promising when compared to the 8–9 cm error margin of the latest camera-based optical systems.
For those interested, Metrica Sports’ contributions are showcased in a GitHub repository and further illustrated in a YouTube presentation by Manchester City’s lead Data Scientist, providing insights into handling their data.
Besides raw data, Metrica Sports equips football professionals with sophisticated analysis tools, catering to the needs of analysts, scouts, and coaches alike.
3.1. Track160
Israeli firm Track160 positions itself as a beacon for mid-sized clubs seeking affordable tracking data and analytical tools. Utilizing a simplified approach involving a single viewpoint camera setup, Track160 aims to make sophisticated tracking solutions accessible to a broader range of football clubs.
The key distinction from conventional tracking systems lies in the significantly reduced technical demands for camera usage, alongside the simplified processes for installation, calibration, and debugging, leading to a noticeable reduction in hardware expenses.
The method for calculating player coordinates has shifted from triangulation in stereo imagery to leveraging machine vision. This innovative approach facilitates broader access to tracking technologies within the football industry.
Furthermore, Track160 enhances its offering with a comprehensive analytics platform, streamlining data processing and the extraction of actionable insights.
For further insights, a presentation by Track160’s CEO at the SPORTBIZ EUROPE 2021 conference is available via the provided link.
4. Summary
In the football industry, tracking data is primarily acquired through three methodologies:
- Optical tracking data employing complex video systems
- LPS tracking utilizing transmitters on players and base stations within the stadium
- GPS tracking based on satellite positioning and specialized trackers on players
Of these, optical tracking stands out for its precision, favored by elite leagues, with coordinate measurement errors approximately around 8–9 cm.
Despite their accuracy, such systems bear a high cost, rendering them inaccessible to smaller clubs and second-tier leagues in certain regions. Innovations in the industry have led to the advent of solutions like SkillCorner, Metrica Sports, and Track160, which capitalize on software to enable tracking through less complex video setups. These firms not only facilitate data acquisition but also offer comprehensive platforms for data management, obviating the necessity for clubs to invest in specialized, high-cost personnel for tracking analytics.
P.S.
Explore further in my blog: a comprehensive guide on generating visualizations from football event data awaits you. Subscribe now to unlock a wealth of knowledge!
