What personal information shows up in a video?

By Check Before Send · Updated

More than faces. A video can show people, vehicle plates and readable text on a screen or sign, it can carry voices, and the file can hold its own details such as where and when it was recorded. The UK regulator says that if individuals can be identified from footage, it is personal data about them.

This guide lists what can give a video away, what the regulators say, and which of these our detectors look for and which they never see.

What the regulators say

The UK Information Commissioner's Office says that if individuals are capable of being identified from surveillance footage, it is personal data about those individuals. It names redaction of both visual and audio data of third parties, and lists blurring, masking or a solid fill as techniques. It also suggests considering removing metadata from footage files.

The European Data Protection Board's guidelines on video devices give licence plates, and information that could identify passers-by, as examples of information relating to a natural person.

What can be in a video

  • People: faces, but also anything else that makes someone recognisable.
  • Vehicles: licence plates.
  • Text in the picture: email addresses, phone numbers, web addresses or codes on a screen, a sign or a document.
  • Sound: voices and spoken names. The UK regulator's page speaks of audio as well as pictures.
  • The file itself: location, date and device, stored in metadata.

What our detectors look for

We built detectors for faces, licence plates and readable on-screen text, and measured them on real and made-up clips, recorded in our own notes.

KindWhat we found
FacesA face in every sampled frame in clips with close and medium faces; one frame in 72 in a dashcam clip where people were tiny
PlatesFound when the frame was split into tiles; trained on Chinese plates, others unmeasured
Text on a screenIn a made-up screen recording, four of five invented targets were found when the frame was enlarged threefold; a code visible for half a second was located, but misread

What they never see

The picture is never understood: only faces, plates and readable text are looked for. Documents on a table, a screen far away, tattoos, house numbers, uniforms and logos are not. Nothing in the sound is inspected; audio is copied across unchanged. And between the frames we look at, nothing is seen: the gap is at most 0.4 seconds.

For videos you export from our page, we do not run a second scan of the picture for faces or plates afterwards. Watching the result is your check. A detector that did not find something has not shown it is not there.

What to do

  1. List what could identify someone in your video: faces, plates, text, voices and the file details.
  2. Cover each yourself where you can, and use a detector as a second pair of eyes.
  3. Watch and listen to the exported copy from start to end before you send it.

Hide sensitive information in a video on this site

Our page runs in your browser and does not upload the video; the Trust Center explains how to confirm that. It finds faces, plates and readable text, lets you pick which to cover, and exports a new file. For the file's own details, see does a video file carry location and date, or browse the other guides.

Sources