01
Treat crowd footage as a tracking problem
Walking footage combines camera motion, changing light, occlusion and people entering from every edge of the frame. A face can appear for only a fraction of a second, then return later at a different scale.
Review should therefore follow the timeline, not only a set of thumbnails. Pay particular attention to crossings, queues, public transport, children and anyone who interacts directly with the camera.
02
Before the edit is locked
Run privacy review before captions, final colour and upload deadlines make changes expensive. Keep an uncensored master with restricted access and create a separate protected export for publishing.
- Scan opening and closing shots.
- Check reflections and shop windows.
- Inspect fast pans and scene cuts.
- Verify the final encoded upload file.
03
When a correction is needed
If detection misses a face, add the region during review and follow it through the relevant shot. If a non-face is selected, remove that track rather than accepting a distracting false blur.
A short, repeatable privacy pass protects the publishing schedule. It is cheaper than taking a live video down, finding the source project and rendering it again under pressure.
Apply the workflow
Review real footage locally.
CensorFlow detects and tracks visible faces on your Windows PC, then lets a person correct and approve the result before export.