Blurring people and vehicles for privacy

Cameras placed for deer or otters will, sooner or later, record a hiker, a ranger, or a car on a track. You didn't deploy the camera to watch people — but once their image is on your storage, it's personal data you're responsible for. Rather than leave that to manual clean-up, the platform handles it at the point of processing.

What gets blurred

The same detection step that finds animals also flags people and vehicles, and blurs both. It removes the whole body, not just the face. That's a deliberate choice: on camera-trap imagery — distant, side-on, from behind, in infrared — a face-only approach would miss too many faces, and a missed face is exactly the failure you're trying to avoid. So the entire detected region is made unrecognizable, with a margin around it so no tell-tale edge of hair or shoulder is left. It runs automatically during processing: nothing to tag by hand, and it's on by default.

Of the two, people are detected more reliably than vehicles — vehicle detection is the newer and weaker side of it — so lean on the person blurring and treat vehicles as the lesser guarantee.

Blurred, not just fuzzed

"Blur" undersells what happens. The region is reduced to a block about a dozen pixels across and then rebuilt from those — the detail is thrown away, not smeared. An ordinary blur is a reversible operation you can partly compute back; discarded pixels can't be recovered. The stored file is overwritten with the blurred version, and every thumbnail and preview is generated from that version — there's no cleaner copy hiding behind it in the app.

The one exception: a 7-day original

There is a single, deliberate exception, and it belongs out in the open. The bar for "this might be a person" is set low on purpose — better to over-blur than to miss someone — which means the system will occasionally blur an animal it wasn't sure about. To keep that recoverable, the untouched original is kept for exactly seven days on separate storage, then permanently deleted by a job that runs every hour. It is not reachable through the application: it isn't tied to any user account or screen, only to whoever administers the storage itself. In plain terms — the original is held for a week so a mistaken blur can be undone, and after that it's gone.

One smaller, honest detail: in the short window between a file arriving and being processed — minutes, not days — it sits unblurred in storage. You can't blur what hasn't been received yet.

Video: the whole clip, and the sound

For video, the entire file is blurred, not just its preview frame, in a single pass. The audio track is removed as well — a recorded voice identifies a person as surely as their face does, so it comes off with the picture.

What it doesn't do

It's a safeguard, not a guarantee, and it's worth being precise about the edges:

• It can miss people. Someone small in the frame, far away, or half hidden behind vegetation can go undetected. Blurring lowers the chance a capture identifies a passer-by; it doesn't promise to catch every one.
• It won't quietly ruin a capture. If a person fills so much of the frame that blurring would leave nothing usable — beyond roughly 40% — the capture isn't blurred. It's flagged instead, so your organization decides what to do with it rather than the system destroying the record on its own.

Turning it on and off

The control lives under Settings → Improve recognition: one switch for the whole organization, changeable only by an organization administrator, and on by default. Turn it off and the platform tells you plainly what that means — that people your cameras capture will be stored and shown as they were photographed. Switching it off only affects footage that arrives afterward; anything already blurred stays blurred, because those pixels are gone.

Data ownership & export → Wildlife AI → ← All knowledge topics