Home / Evidence redaction
Redaction for Evidence
One incident produces footage from several cameras, a dispatch recording and a report, and all of it is potentially releasable. Volume is the constraint, not difficulty.
Use Cases
Body Camera Footage
Video
Detect and mask faces, plates, weapons and spoken PII across body-worn camera footage. Built for records teams handling high-volume public records requests.
Learn moreCCTV and Surveillance
Video
Blur faces and identifiable objects across hours of continuous footage from any camera system. Multi-camera footage processed in bulk.
Learn moreDashcam and In-Vehicle
Video
Redact faces, license plates and identifiable objects captured in dashcam recordings. Used by fleet operators, insurers and law enforcement.
Learn more911 and Dispatch Calls
Audio
Redact caller identities, location details, medical information and case-specific data from emergency dispatch recordings before release or disclosure.
Learn moreFace Redaction
Video
Detect, track and mask every face in video and images. Handles crowds, movement, occlusion and low light.
Learn moreLicense Plate Redaction
Video
Detect and blur license plates across video and images. Used in surveillance footage, fleet recordings and public release video.
Learn moreWhat These Share
Faces move, frame to frame
At 30 frames per second a minute of footage is 1,800 decisions about the same person. Detection runs per frame and tracking carries the mask through movement, turning away and re-entry.
The backlog is the problem
Queue-based bulk processing runs the same policy across a folder of files, permissioned per format, rather than one operator opening one file at a time.
Custody holds end to end
Roughly 36 event types are recorded, with the user, their email address, the IP address they acted from and the timestamp on each.
It cannot leave the network
Criminal-justice material is routinely subject to a boundary requirement, which is why the entire AI stack runs on-premises or air-gapped when it has to.
From Ingest to Release
Ingest
Footage, dispatch audio and stills from every camera on the incident land in one place, hashed on arrival so integrity can be shown later.
Detect and track
Faces, licence plates, weapons, screens and street signs are detected per frame, and tracking carries each mask through movement, occlusion and re-entry.
Review
An operator scrubs the timeline, corrects a tracked object where it drifts, and adds anything the model did not raise.
Apply
Blur, pixelate or solid fill is burned into a new rendition, with the exemption drawn onto the mask where the release requires one.
Release
The redacted copy goes out and the original stays, because an appeal against a withholding is decided against the unredacted record.
Where This Fits
Most evidence work is video, and video redaction covers tracking, 360 degree footage and the formats body-worn and in-car systems produce. Dispatch and interview audio is audio redaction. Automation and scale covers queue-based bulk processing for a backlog, and deployment covers running the whole stack inside your own network when criminal-justice policy requires it.
FAQ
Questions, answered
What evidence types does this cover?
Body-worn camera footage, CCTV and surveillance, dashcam and in-vehicle recordings, 911 and dispatch audio, and the faces and licence plates that appear across all of them.
How is a moving subject tracked across a recording?
Detection runs per frame and tracking carries the mask through movement, turning away, occlusion and re-entry. At 30 frames per second a single minute is 1,800 decisions about the same person, which is why tracking rather than per-frame drawing is what makes footage tractable.
Can a backlog be worked through in bulk?
Yes. Queue-based bulk processing runs the same policy across a folder of files, permissioned per format, rather than an operator opening one file at a time.
Can evidence be redacted without leaving our network?
Yes. Criminal justice material is routinely subject to a boundary requirement, and the entire AI stack runs on-premises or air-gapped when it has to.
How is chain of custody maintained?
Roughly 36 event types are recorded from ingest to disposition, each with the user, their email address, the IP address they acted from and the timestamp.
Send Us an Hour of Footage
We will redact it and show you the custody trail alongside the output.