Face blurring - exact legal obligations? (in GDPR-land)


This topic has come up once or twice already but thought I’d raise it again in the light of experiments I’ve made with both pii-blur (TrekView) and the understand.ai anonymizer. I’ve had variable results with the former but quite encouraging results with the latter.

With the understand.ai anonymizer (https://github.com/understand-ai/anonymizer) I get pretty good results for faces that are clearly visible. Don’t have too many panos with license plates so untested as yet, but one with a vaguely-visible license plate (which you can’t read properly even if unblurred) was not detecting.

However I have a question on our legal obligations (my server is in Germany, I am in UK, so both in GDPR-land - though the software will potentially be deployed anywhere in the world - this is Eesger and myself’s OpenWanderer project which he’s mentioned elsewhere): to what extent do we need to blur faces and license plates?

With the understand.ai tool I can blur faces that are clearly visible, and the TrekView pii-blur tool adds further blurring on entire people and cars though it doesn’t detect as well as understand.ai.

However, faces further away from the camera are not reliably detected and blurred - but these faces are not clearly visible anyway. Do we thus have to blur ALL people showing on the panorama irrespective of whether their face is clearly visible, or just faces that are clearly visible? From a privacy and ethical POV I’d have thought just clearly visible faces, but IANAL.

Intiial experiments here incidentally


1 Like

Have a read of this @nickw

Our data protection regime lets us take holiday snaps, even of strangers, provided we’re doing so for private purposes. But if we’re taking snaps for commercial use, where individuals are identifiable, there is no such exemption.

Source: https://www.pinsentmasons.com/out-law/news/googles-street-view-could-be-unlawful-in-europe

This is dated before GDPR, but it seems anything in the public domain that could be construed as commercial use (I’d argue a dedicate street-level platform is commercially orientated) is a grey area.

Speaking with no legal experience, it seems from services that exist, showing you are making a best effort to remove personally identifiable information might be enough, but it’s clearly a grey area, and one I’d advise seeking proper legal advice on.

1 Like

To follow up on this, I had a pointer from the OSM mailing list to another face blurring tool, one which appears to work very well due to splitting each pano into rectangular segments and then detecting faces and cars in each segment.

See https://github.com/tyndare/blur-persons/

I’ve tested this on OTV panos and it appears to work very well. There was just one case of undetected faces close-ish to the camera, in which two children partly looking away from the camera were not detected. They were on the edge of one of the segments which could explain things?

It takes longer than some of the alternatives (5 min per pano typically) but it works well so I will probably use this for OTV; @Eesger this looks good for OpenWanderer.

See for example
https://opentrailview.org/?id=10015 (car)
https://opentrailview.org/?id=9728, https://opentrailview.org/?id=9730 (people)
https://opentrailview.org/?id=3434 (car, close)

Note that people in the distance are not detected, but I suspect we don’t need to worry so much: e.g.:

HQ | Blog | Spotted a Trekker? | Become a Trekker | Facebook | Instagram