Video and Body-Worn Camera Redaction

Video redaction means tracking and obscuring faces, screens,documents, and identifying audio across every frame in which they appear. Anhour of footage can take many hours to review, which is why video requestsdominate the backlog in agencies that receive them.

The arithmetic is what surprises people. A document requestthat produces two hundred pages is a large request. A single body-worn cameraactivation that runs forty minutes can consume more staff time than those twohundred pages, and a use-of-force incident may involve six cameras recordingthe same forty minutes from six angles.

A redaction on a page is a decision made once. You identifythe material, you obscure it, and the page is finished.

Video is not one decision. It is a decision that has to holdacross every frame in which the subject appears, while the subject moves, turnsaway, is briefly occluded by a doorway, and comes back. A face that iscorrectly blurred for nine seconds and visible in the tenth is not a partiallysuccessful redaction. It is a failed one, and the ten seconds are what willcirculate.

Audio compounds it. A bystander's name spoken once has to belocated and muted, and names are spoken in the middle of sentences, over radiotraffic, at conversational speed. Reading a transcript is far faster thanlistening, which is why transcription is usually the first thing worth doing toa video, before any redaction decision is made at all.

The categories are familiar from documents, but they appearin unfamiliar places.

Faces, of bystanders, victims, minors, and sometimesofficers, depending on the jurisdiction and the nature of the incident.

Screens and documents caught incidentally. A computermonitor in the background of a booking area, a driver's license held up to thecamera, a form on a desk. These are easy to miss because nobody is looking atthem; they are scenery.

Identifying audio. Names, addresses, phone numbers,dates of birth, and medical information, spoken by anyone on the recordingincluding the officer.

Location detail where the address itself isprotected, such as a shelter or a residence in a domestic matter.

Reflections. A face in a car window or a mirror isstill a face, and this is the category most often missed, because the revieweris watching the subject rather than the surfaces around them.

Not, mostly, on applying the blur. Modern tooling can tracka face through a scene reasonably well once it has been told which face totrack.

The time goes on the first pass: watching the footage tofind out what is in it. Someone has to know that a bystander appears at 12:04,that a monitor is visible from 18:30 to 19:10, and that a name is spoken at26:45. Until that inventory exists, nothing can be redacted, and building it isa real-time activity. Ten hours of footage takes ten hours to watch, atminimum, and more if it is reviewed carefully.

That is the constraint worth explaining to leadership whenthey ask why a video request is taking so long. It is not slow tooling. It isthat the material has to be observed before it can be judged.

Before any of this, someone has to decide whether thefootage is releasable at all, and in what form.

That is a legal question that varies considerably byjurisdiction and by circumstance: whether an investigation is active, whetherthe recording was made inside a residence, whether a minor is involved, whethera death occurred. Some states have written body-worn camera provisions directlyinto their public records statute; others handle it through the general lawenforcement exemption.

The practical point is that the redaction work should notbegin until that determination is made, because redacting sixty minutes offootage that ultimately gets withheld in full is the most expensive way toreach a decision.

Govflo redactsvideo, and transcribes and summarizes body-worn camera footage so the firstpass can be read rather than watched. The transcript makes the inventory stepsearchable: instead of watching for the moment a name is spoken, you find it intext and go to the timestamp. Where the audio is poor enough that thetranscription is uncertain, those segments are set aside for a person tocorrect rather than passed through as though they were reliable.

Detection proposes the regions to obscure, faces and platesin the picture and the segments to silence in the audio, and a caseworkerconfirms them. It is the same assist-then-approve pattern used on documents,and files at the scale video actually arrives in, into the hundreds ofgigabytes, are handled as uploads rather than as an obstacle.

What it does not do is make the exemption call. Whetherfootage is releasable, and in what form, is a legal determination that stayswith the agency and its counsel. The tooling changes how long the work takes,not who is responsible for the judgment.

Transcribe first, always. Even when you intend towatch the whole thing, having the transcript alongside makes the second passmuch faster and gives you something searchable for the appeal.

Log timestamps as you go. A list of what appears whenis reusable if the same footage is requested again, which it frequently is.

Handle multi-camera incidents as one job. Six anglesof one event share most of their redaction decisions. Working them separatelymeans making the same call six times, inconsistently.

Talk to the requester about scope. Many videorequests are satisfied by a specific window rather than the full activation.Asking is faster than redacting an hour to deliver four minutes of relevance.

‍