The Wingfield Court is designed for both indoor and outdoor use. To be prepared for wind and weather, particularly robust materials were manufactured. The court is approved by the International Tennis Federation (ITF) and therefore meets the criteria for Player Analysis Technology (PAT).

The heart of the court is the Wingfield Box. It is not only a point of interaction for us players, but primarily a data aggregator. Its two integrated high-speed cameras capture everything that happens on the court and collect moving image data of our match.

But how do we get from simple video recordings to serve rates or shot speeds? How does the system know the score or which rally is worth recording in a highlight clip? It’s all about the evaluation and interpretation of this “raw” image data. And this is where the software comes in: our AI “Walter” - Wingfield’s brain - named after our namesake Walter Clopton Wingfield, who once brought tennis to the masses.

So to let our system “understand” what it “sees” through the cameras on the court, we use computer vision and AI processes that are also used in autonomous driving, for example. Our software can be divided into four building blocks: Human Recognition, Ball Recognition, Shot Recognition, and Tennis Logic.

First and foremost, the goal is to extract as much tennis-relevant information as possible from the image. Using machine learning techniques, we have taught “Walter” what a person or a ball is, for example. This way, we understand at any given moment where a player is on the court and what he is doing: is he running or taking a break? Is he a coach or more of a player? Is he playing a forehand or a backhand?

We add tennis logic to all of this collected raw information. For example, a ball is now no longer just somewhere in an undefined space, but inside or outside the defined court. Also, we now know if a point was just won or how fast the hardest serve was in a match.

A great feature of our match mode is the automatic scoring. “Walter” counts the result in the background. After all, we want to be able to call up video sequences and statistics for the appropriate scores at the end of each match.

This is where the tennis logic comes into play again. Based on the information of tracked trajectories, player movements or shots, “Walter” detects certain tennis specific events (e.g. serves or side changes).

To make sure that the tracked match statistics match your score, “Walter” reacts to your behavior on the court. He recognizes how you and your partner count. If he has once decided differently than you in a situation, for example in the case of a close ball that you have mistakenly called OUT, he analyzes the subsequent situations and compares them permanently with his own decisions. In case of doubt, he corrects himself, even if the decision may have been objectively wrong. Because what counts in the end is getting the right statistics for the right result.

The good thing about “Walter” is that he continues to learn and gets better and better over time. The big challenge, however, remains the handling of previously unknown situations. To give you a better idea, here are two anecdotes from the past:

In our early days, it sometimes happened that the serves of some of our very young players were not recognized. “Walter” had already learned how a serve should look like, but the data sets used for training so far only included serves of adult players. Reason enough to order “Walter” to do some extra training.

During a match at the Road to Wimbledon junior tournament, over 30 birds flew up in the background of the court. Our software recognized these as balls at first, resulting in several thousand potential trajectories being tracked. To avoid such scenarios in the future, we are developing various interception mechanisms, so we have now also tamed the birds in London.