anon

Anonymizing videos with face blurring, logo blurring, and speech redaction.

Author: @snlee159

Acknowledgements

This project was made in collaboration with @FARLab to meet some of their needs for automatic face blurring for showing participant study research publicly.

Purpose and Context

Oftentimes videos need to be shared publicly for demonstrating a technology or project. However, it is critical to protect a participant's identity.

Anon anonymizes videos for this purpose for blurring faces, redacting proper nouns from audio, and blurring logos on shirts.

How to Install

Only available for developer installation right now. Install via git for now with:

git clone https://github.com/snlee159/anon.git
cd anon
sudo python setup.py

to install all requirements.

You'll also need to install the vosk model for speech redaction. You can do that with the following:

cd anon
cd vid_anon
wget https://alphacephei.com/vosk/models/vosk-model-en-us-0.22.zip
sudo apt-get install unzip
unzip vosk-model-en-us-0.22.zip
rm vosk-model-en-us-0.22.zip

You should now be ready to run the program.

How to Use

Command Line Interface

Anon must be run via a command line argument (within the anon directory) with the following additional arguments:

python vid_anon <video_directory> [optional arguments]

Where video_directory is the path to the directory where every video you want to anonymize is located. Any file ending in .MP4 or .mp4 in that directory will be anonymized. Do not include the trailing backslash at the end of the video directory path.

Optional arguments include:

• --blur_face: to automatically blur faces in the video; will blur all faces in the videos
• --blur_shirt: to automatically blur logos on shirts. All shirts will be blurred
• --redact_names: to remove proper nouns from the audio
• --cut_first_last_mile <drive_times>.csv: to remove the first and last mile from drive start/stop time in a video which includes driving. A lot of FARLab's videos including in-vehicle recordings and this was included to protect participant home location information. Requires the drive_times csv path to the file to be referenced. Which includes the following columns: filename, start_frame, end_frame where filename is the exact filenames (with .mp4) in the video_directory and start_frame and end_frame are the frames, not the times, when the driver starts/stops driving. Assumes a video of 30 fps
• --clean_up remove support files (audio chunks, etc)
• --output_dir <output_directory>: default will save the anonymized files to where this is run. If this is set, output_directory will be the path where the anonymized files (and support files, if not deleted) will be stored (include the trailing backslash in the output directory path).

Exceptions and Things That May Break

Right now, this only works properly for .mp4 or .MP4 file extensions.

In addition, videos over 10 minutes may experience issues.

Future Improvements

• Make intermediary video files to avoid memory overload that breaks process for long videos
• Save face and body points to CSV for reference post-anonymization
• Remove trailing backslash for path specification for easier command line arguments