How to blur faces in 360° street view images
Anyone publishing street-level 360° imagery has the same obligation: the people captured in those panoramas must not be identifiable. This guide walks through a reliable workflow for detecting and blurring faces across equirectangular images — from a single panorama to a full survey of thousands.
Why faces in 360° images are harder to blur
An equirectangular panorama is a sphere stretched into a flat 2:1 rectangle. Near the left and right edges, a single face can be split across the seam; near the poles it is heavily distorted. A flat-image face detector that scans the rectangle once will miss many of these. Reliable anonymisation requires detection that understands the projection and works at multiple scales, so small or distorted faces are still caught.
Step-by-step workflow
The workflow below is how Privacy Keeper handles it, but the principles apply to any tool you evaluate.
- Load your images. Point the tool at the folder of equirectangular JPGs straight from your camera — Mosaic 51, Insta360 X3/X4/X5, GoPro MAX, and generic 360 output up to 32K are all supported.
- Run AI face detection. The detector scans each panorama at multiple field-of-view scales, projects detections back onto the sphere, and removes duplicates so every face is found once. A 12K Mosaic 51 panorama processes in roughly 9 seconds on an RTX 3090, with an automatic CPU fallback when no GPU is present.
- Review every detection. Open each image at full resolution and confirm, add, or remove blur regions by hand. This is the difference between "mostly anonymised" and audit-ready.
- Export. Anonymised JPGs are written out while the originals stay untouched, and EXIF, GPS, and other metadata are preserved so your imagery still aligns on the map.
Processing at scale
Survey work means thousands of panoramas, not one. Batch processing runs the whole folder unattended, and checkpoint/resume means a long job can stop and restart without reprocessing what is already done. Every run writes an audit log recording system details, settings, and per-image results — useful evidence that anonymisation actually happened.
Keep it offline
Faces are personal data. Uploading them to a cloud service to blur them creates exactly the exposure you are trying to remove. On-device processing keeps every image on your own hardware, with no telemetry and no cloud upload — see our guide on GDPR compliance for street-level imagery for why that matters legally.