Learning goal
After watching, you can explain the privacy–utility trade-off in your own research data, name the main types of identifying information in a video, and decide whether masking is relevant for your next study.
Transcript
Cold open
A developmental psychologist sits on hundreds of hours of parent–child interactions. The data is gold. The data is also locked in a drawer. Nobody else gets to look at it. It can't be replicated, archived, or reused. This happens everywhere. And it's the problem video masking exists to fix.
The privacy–sharing tension
Every privacy decision in video research is a trade. The things that identify a person — face, voice, gait — overlap with the things you want to measure: emotion, attention, gesture, posture. You can't strip one without hurting the other.
There's no single answer. Heavy blur protects privacy but kills gaze tracking. A silhouette preserves movement but loses facial expression. A skeleton overlay strips almost everything visual but keeps body kinematics. The right choice depends on your research question.
Meanwhile, the rules are tightening. GDPR. Ethics boards. Funder data-sharing mandates. The push to share is colliding with the push to protect. Masking is what lets you say yes to both.
Five types of identifying information
When people think privacy in video, they think faces. But there are five buckets, not one.
Faces — the obvious one. Structure, expressions, distinctive features. Even partial visibility leaks information.
Body characteristics — height, gait, posture. Gait alone can identify someone almost as reliably as a face. Clothing too, especially anything distinctive — a uniform, a piece of jewellery.
Behavioural patterns — gesture style, speech rhythm, how someone interacts. These persist even when the visual is gone. Voice on its own is identifying.
Context — the background. Buildings, posters, clocks, other people who walk through frame. Anything that could be cross-referenced.
Temporal patterns — when the recording was made, the EXIF metadata, seasonal cues, recurring schedules across sessions.
Real masking addresses all five. Not just the face.
When masking is worth your time
You should consider masking your video data if any of these are true:
- You're sharing with collaborators — even ones at your own institution.
- Your institution or funder requires a privacy plan.
- You want to archive the data for future researchers.
- You want the data to be useful for secondary analyses you haven't thought of yet.
- You're building or using machine-learning models, and need the data outside a locked environment.
- You've already hit a wall: someone said no, the data sits unused, you don't know where to go.
If any of these are true, the rest of this tutorial series is for you.
Recap
Three things to remember. One: every masking choice is a privacy–utility trade. There's no free option. Two: identifying information is five buckets — face, body, behaviour, context, time — and good masking touches all of them. Three: if you're sitting on data you can't share, masking isn't a luxury, it's the unlock.
Next up: getting the tool installed. We'll get MaskAnyone running on your machine in under five minutes.