How to reconstruct an incident from CCTV

Done by hand, reconstruction goes wrong in a predictable way. You find the incident itself, eventually, and then the real work starts: which camera saw the ten minutes before, where did that person come from, what happened just off the edge of frame. Each new question is another scrubbing session, and by the third one the details are being argued from memory rather than checked. The incident took thirty seconds. The reconstruction takes days.

An investigation is a list of questions

When did the pallet fall. Who was in the aisle before it. Where did the forklift go next. Every reconstruction reduces to questions like these, and the reason they're painful is that each one traditionally means watching more video. Once a vision model has described every frame, each question becomes a plain-English search instead of a viewing, with results in seconds. A working theory can be tested in minutes, rather than defended in a meeting because nobody has time to check it.

1. 01

   Pin the incident

   Search for the event itself by description: “someone slipped in the entrance”, “forklift reversing without a spotter”. The match fixes the time and camera for you, which is the anchor everything else hangs off.

2. 02

   Work backwards

   Search the window before the incident, on the same camera and its neighbours, for causes and context. What was delivered, who walked through, what changed. Forwards works the same way for what happened after.

3. 03

   Find the people by description

   Run the same description over each camera's footage: “man in a grey coat” at 15:42 on the dock camera and 15:47 in aisle three is an order of events you've established by searching, not by tracking anyone. Every match carries its timestamp.

4. 04

   Assemble the sequence

   Order the moments you've found into a timeline. Incident Reconstruction, one of the report checks, reads a video's whole timeline and writes it up in order, so the output is a document rather than a folder of clips.

Across cameras is where it earns its keep

Incidents almost never sit politely on one camera. The cause is on the loading dock camera, the effect is in aisle three, and the person who connects them passed four other lenses in between. Keep the relevant footage in one folder of the Library and the search bar covers all of it in a single query, subfolders included, so the same description surfaces every sighting with its time. Svid doesn't link those sightings to an identity or follow anyone between lenses, deliberately; you establish the order from the timestamps, which is also the version that holds up when someone asks how you know.

Recorded or live, exports included

This works on the recorder's live feed or on exports pulled after the fact, because live and recorded footage go through the same pipeline. A pile of clips exported onto a USB stick is a perfectly good starting point; upload them and ask the same questions. On cost: the platform is priced from your own usage with a $9 a month minimum, and report checks like Incident Reconstruction start at a $3 a month floor, priced from your own usage in the Store. If the investigation is sensitive, a fully on-prem appliance mode exists, so the footage never leaves the building. And if the destination is an insurance claim, the same method applies with a bit more paperwork around it.

How do I find when something happened if I don't know the time?

Search for the event itself by description rather than by timestamp. “Someone slipped in the entrance” finds the moment, and the moment tells you the time, not the other way round.

Can I follow one person through the whole incident?

Not automatically, and that's deliberate: Svid doesn't track people or link sightings into an identity. What you can do is run the same description over each camera's footage and order the timestamped matches, which establishes the route from evidence you can point at.

Does this work on footage exported from my recorder?

Yes. Uploads go through the same pipeline as live feeds, so exported files from any recorder can be read, searched and assembled into a timeline the same way.

What if the incident spans several cameras?

That's the normal case, not the exception. Keep the footage in one folder and a single search covers all of it, subfolders included; the order of events comes from the timestamps on the matches, and Incident Reconstruction writes each video's timeline up for you.