This is an experimental application to obfuscate faces and license plates in OpenWanderer projects. Note that it makes use of these existing anonymisers:
Both these libraries are included in this repository for easy installation; I have also modified their code slightly to make it work with this application. The code from these repositories was taken in autumn 2020 (as at Feb 2022, this remains the most recent version for both).
Please see the individual licensing for the two anonymisers. The rest of the code is the OpenWanderer project's own code and is licensed under the LGPL.
It finds all unauthorised panoramas in an OpenWanderer database, and attempts to anonymise them, using one of the above anonymisers. Optionally, you can specify a range of pano IDs in the database to anonymise.
(Taken from the instructions for understand.ai Anonymizer and modified)
To install the anonymizer just clone this repository, create a new Python virtual environment and install the dependencies.
The sequence of commands to do all this is
python3 -m venv ~/.virtualenvs/anonymizer
source ~/.virtualenvs/anonymizer/bin/activate
git clone https://github.com/openwanderer/anon.git
pip3 install -r requirements.txt
python3 anon.py [-a anonymiser] [-f startPanoID] [-t endPanoID] [-a anonymizer] -i InputImageDirectory -o OutputImageDirectory
- The
anonymiser
should be eitherunderstandai
(default) orblur_persons
. startPanoID
andendPanoID
define the range of panorama IDs in the database to anonymise. If not specified, all unauthorised panoramas will be anonymised.InputImageDirectory
is the location of the unanonymised images.OutputImageDirectory
is the location to write out the anonymised images
Of the two, blur_persons
is the most effective, anonymising almost all visible faces and cars. However it blurs the whole car, not just the license plate - which is not ideal, and critically, it is very slow, taking several minutes to anonymise one panorama.
understandai
is much faster, taking a few seconds to anonymise panoramas, and does work on the majority of cases though might miss faces and license plates which are some distance away from the camera. It does only blur the license plate, not the whole car, which is an advantage.