Research in Focus: Make your own goalkeeper data

'Learning From the Pros: Extracting Professional Goalkeeper Technique from Broadcast Footage', by Matthew Wear, Ryan Beal, Tim Matthews, Tim J. Norman, and Sarvapali D. Ramchurn

[Link to the paper is here]

Why it's worth your time

Goalkeeper analysis might be the most overlooked area of them all, so anything in that area is worth checking out. But this paper is also an example of how you can find some really great and unique data by yourself.

What it says

The paper starts off by collecting data from TV footage, specifically the still images of a goalkeeper's saves (or failed saves if they concede).

The images were taken when the striker made contact with the ball in 1v1s or penalties, from Premier League and 2018 World Cup matches. These images were then run through body pose estimation software (PoseHG3D), giving you neat stick-figure data.

This body pose data (as well as a 'goalkeeper engagement metric') is classified into save types using k-means clustering, which can then be used for analysis. For example, the researchers looked into which save techniques appeared to be the best options in different situations, which has obvious applications.

What's cool about it

I'm personally really excited about the possibilities that body pose data brings. However, it's always seemed out-of-reach to me, but the fact that this paper derived it from TV footage brought it back to reality.

Part of this is because this is a smart question to ask, and one of the few things in football where (between live action and replays) you'd reliably be able to get a good enough still image to derive the data. Maybe you'd be able to flip it and do something for ball-striking technique/finishing choice too.

Having more analysis of goalkeepers is also useful. Four clusters was found to be optimal by the researchers, with them naming the groups: aggressive set, passive set, spread, and smother (representative images of the groups are available in the paper). This kind of categorisation/conceptualisation can be built on and used as starting points for future research. For us to stand on the shoulders of giants we need those shoulders to be created in the first place.

'Research in Focus' is like SparkNotes for football analytics: summarising and analysing the best research out there. Follow this link for the list of all Research in Focus pieces.