Face Blurring
A support call puts a camera on someone's head in a workplace full of other people. Face blurring covers the people who walk through the frame. A call about a broken pump does not become a record of everyone on the floor.
Turn it on and faces are blurred on the fieldworker's phone, before the video is sent. The expert never receives an unblurred frame. It works on photos too.
Available since version 2.1.4 on Android and iOS. Face blurring needs a license. See licensing.
Turning it on
There are three ways:
- The face blurring setting in the app
- The face blurring button during a call
- By voice. Say "Enable face blurring" or "Disable face blurring". The command works in every supported language. See voice commands.
The expert sees when blurring is active, so both sides know what is being sent.
Where it applies
| Blurred | |
|---|---|
| Live video to the expert | Yes |
| Photos taken during the call | Yes |
| The fieldworker's own display | No. They see the real scene. |
How it works
Detection runs on the device. Android uses Google ML Kit. iOS uses Apple Vision. No frame is sent anywhere to be analysed. Nothing about a face is stored or transmitted.
The detector only returns the position of a face. Landmark detection, contour detection, and classification are switched off. Assist finds where a face is and nothing else. It does not build a face template. It cannot recognise or match a person. It cannot tell you whether the same face turned up in two calls.
Faces are found in any orientation, including upside down. That matters when someone looks up into a machine or down into a pit.
Not a biometric system
Face blurring finds where faces are so it can cover them. It does not identify anyone. It produces no biometric data under GDPR Article 9. If your works council or DPO asks, that is the answer.
Limits worth knowing
Face blurring reduces how often bystanders are captured. It is not a guarantee. Do not present it to staff as one.
- Small faces can be missed. A face smaller than about 5% of the frame width may not be found. In practice, that means people well in the background.
- Detection samples the video. It runs about fifteen times a second, on a smaller copy of the frame.
- Fast movement is covered by margin. Detected areas are made about a quarter wider to allow for movement. That is why the blurred patch looks bigger than the face.
- Heavy blocking defeats it. A face mostly hidden behind equipment, a mask, or an arm may not register.
If a site requires that no bystander is ever captured, use face blurring as one layer. Keep the controls you already have: restricted filming zones, notices, and briefing the people working nearby.