EURO 2024: How player tracking works technically

Page 4: Virtual slow motion in 3D

Contents

Multi-person real-time 2D algorithms such as OpenPose are at the heart of numerous applications in sport - including largely automated offside detection. However, because these are statistical models that have developed their capabilities based on training data, it is important to be aware of their system-related weaknesses: Particularly in controversial, turbulent scenes where players are operating hard on the edge of the offside rule and obscuring each other, limb recognition can also fail: for example, incorrectly assigning a foot or misjudging its position.

Empfohlener redaktioneller Inhalt

Mit Ihrer Zustimmmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.

Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.

Unfavorable lighting conditions can also influence the recognition rate, as can distortions in the training data: Strictly speaking, these are prediction systems and not recognition systems. Because several camera perspectives are analysed and combined to create an overall image, sources of error can be minimized but not eliminated.

The players on the pitch may not even notice many of the system's misjudgments. This is because when the SAOT triggers an alarm, the assistant referees first check on the monitor whether the automatically calculated lines and markings are plausible. Only if the result is positive is the referee on the pitch informed and the game interrupted.

However, despite the improved technology and the fact that checks are largely carried out in the background, there are still unclear scenes that are difficult to assess and cannot be decided within 15 to 30 seconds. There should therefore be enough time and material for lively discussions.

(atr)