Camera analytics that turn footage into an event list
- One row per event
- Timestamps you can jump to
- Your footage stays on your machine
Interior, after hours
Fixed camera, low light, two subjects
Analysing the clip
Clip
Decoded in your browser. Your video is never uploaded.
Sets the luma delta threshold and the smallest region that counts.
Kills the one frame flicker that no operator wants paged about.
Exclude zone
Put it over the road, the tree, the neighbour's window. Events inside it stop alerting.
The frame an event was found on shows up here.
Nothing selected yet
Hover a row in the event table to preview its frame, click it to hold the frame and jump the clip there.
- Length
- 0.0s
- Confidence
- 0%
- Area
- 0.0%
0
Motion events found
0
Dropped by zone
0
Dropped, too short
0
Alerts you would get
| # | Start | Length | Class | Conf. | Area |
|---|
Reading the clip. Events appear here as they are found.
Get these alerts from your own cameras.
The same incident, reviewed two ways.
Without analytics
Scrub, guess, scrub again
- Somebody reports a unit was opened overnight.
- You open the recorder and pick a camera.
- You scrub at eight times speed from 22:00 and watch an empty aisle.
- You miss it, go back, and watch it again at four times.
- Forty minutes later you find 01:47, and now you need the gate camera for the same window.
- You repeat the whole process on camera two.
With an event timeline
Read seven rows, open one
A worked example of the output shape, written to show what a night looks like once events are classified. It is not a customer's footage and not a measured result.
Four parts, and each one answers a different question.
This is the same object the demo produces on the homepage, described rather than drawn, because it is the thing you would actually be buying.
The scrubber
A tick under the video at every event position, coloured by class. One glance says how the night was distributed: a cluster at 02:00 reads differently from an even scatter across eight hours.
The event table
One row per event, dense on purpose. Start time, duration, class, confidence, motion area and a one line description. Sparse tables look like a mockup, dense ones look like software.
The bounding boxes
Drawn on the frame as each event window comes up, so you can see what the detector actually locked onto rather than trusting a label.
The summary strip
Events found, dropped by zone, dropped by duration, kept. The dropped counts are the argument: they are the alerts that would have reached your phone and now do not.
Every figure in it is measured rather than styled: timings come from frame indices, motion area from the changed region, confidence from the model itself. Nothing is rounded up to look better.
This category oversells, so here is the boundary in advance.
Not identity
No face recognition, no face templates, no re identification of the same person across cameras or across days. The timeline says a person was there. Who it was is your call, made with your access log.
Not counting people for marketing
No footfall dashboards, no dwell time heat maps, no queue analytics. Those are a different product with different privacy consequences, and building them would compromise the sentence above.
Not prediction
Nothing here scores behaviour as suspicious or predicts intent. A model that infers intent from posture is a model that will be wrong about a person, and there is no version of that we want to ship.
Not a recorder
We hold events, not footage. Your video stays where it is now, on your recorder or your machine, and retention of the video remains your policy and your storage.
The full version of these commitments, in the wording a security review expects, is on the enterprise security camera systems page.
Three moments, and only one of them is an emergency.
Overnight
Person and vehicle events reach a phone. Animal and other motion stay in the timeline where they belong. If nothing happened, nothing arrives, which is the whole point.
The morning check
Two minutes reading last night as a list. Most mornings this ends with nothing to do, and knowing that quickly is worth as much as catching something.
After an incident
Filter to the window, read the sequence, export the clips that matter. What used to be an hour of scrubbing is a couple of minutes, and the timestamps are precise enough to hand to somebody else.
The four verticals this is used on, and the specific noise each one fights, are on the video analytics platform use cases hub.
Four practical questions.
01
Do you store our video?
No. We hold events and their metadata, not footage. In the demo the video is decoded in your browser and never uploaded at all, and only small cropped keyframes of detected events are sent for classification.
02
How long are events kept?
Seven days on Watch, thirty on Operate, ninety on Multi-Site, and a custom retention policy on Enterprise. Every tier and its retention is on the pricing page.
03
Can I export the timeline?
Yes. Events are data, not pictures of data, so a filtered timeline can be exported for an insurer, a landlord or a police report with the timestamps intact.
04
What does confidence actually mean?
It is the model's own value for that classification, shown exactly as it came back. A low number is displayed as a low number rather than being hidden, because an analytics product that only shows you its confident answers is not an analytics product.
Related pages on reading footage
Video management software for IP cameras
Six VMS platforms on licence model and price, and which half of the job you already own.
CCTV video analytics software
The same timeline on an older ONVIF or DVR estate.
Motion detection software
The detection stage that produces the events.
AI video analytics software features
Every capability, grouped by the job it does.
Warehouse video surveillance
Docks, forklifts and the yard between shifts.
Video analytics tools
Five classes of tool and what each really detects.
Video analytics software price
What the timeline costs per camera per month.
Turn one night of footage into one screen of rows.
The demo does it on a bundled clip or on your own, in your browser. When you want it running on your cameras, one email address starts it.