How blurring works

The question club boards ask most is what the sensors actually see and what is stored. Here is the answer, step by step.

1. The sensor watches the playing area

Sensors sit at selected locations and face the playing areas. The sensor sends an encrypted video stream to TeeGuard's service in the region, where every frame is analysed in memory.

2. The service only looks for people

The model in the service is an object detector set to a single class: person. It draws a box around each person it detects. It has no function for finding faces or reading body posture, and it cannot tell who anyone is. Our codebase has an automated check that stops the build if anyone adds known face or pose detection libraries.

3. The head is blurred

For each person it detects, the service blurs the top 30 per cent of the box, where the head is, with a heavy blur. This happens in memory, before anything is stored in the cloud. We blur a fixed part of the box rather than search for faces, because searching for faces would need exactly the face detection we refuse to use.

4. What is stored

• The number of people in the group.
• The time.
• Clothing colour, top and trousers, used to recognise the same group at the next sensor.
• Height band: a 10 cm range from size in the image, never an exact height.
• One blurred image per event.

The video stream is analysed in the service's memory and never stored. If the sensor loses contact, nothing is analysed and no image is stored. No video is recorded.

5. How long it is kept

The blurred image is deleted automatically and is never kept for more than 24 hours. Metadata is kept for 30 days unless the club has agreed otherwise. The database and the images are stored in the EU, and each club sees only its own data.

Blurred images still count as personal data under the GDPR, because in some cases it may be possible to work out who is in them. So we treat them as personal data throughout: under contract, with access control and with deletion.