AI security camera software for the IP cameras you have
- No new cameras
- No appliance on site
- Published price, no sales call
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.
If your camera speaks ONVIF or RTSP, it works, and almost all of them do.
This is a compatibility question, not an AI question. The model never touches your camera, it looks at frames the camera already publishes.
Hikvision, Dahua, and the brands built on them
Uniview, Annke, Amcrest, LTS and most of the white label market. RTSP is on by default on nearly all of them.
Reolink
Main and sub streams both published. Use the sub stream for detection and keep the main stream for recording.
Axis, Bosch, Hanwha, Vivotek
Full ONVIF profiles. These are the easiest cameras in the category to work with.
Ubiquiti UniFi Protect
Enable the RTSP stream per camera in the Protect interface, then use the URL it gives you.
Analogue cameras on a hybrid DVR
If the recorder publishes an RTSP stream, and most made in the last decade do, the cameras themselves can be twenty years old.
Consumer cloud cameras
This is the one honest no. If a camera only speaks to its manufacturer's app and publishes no stream, nothing external can read it, including us.
Finding the stream URL for your particular model is the only fiddly part, and it is written out per brand on the RTSP software page.
Two stages, and only the second one is a model.
Stage one, arithmetic
Frame differencing against a rolling background, thresholded, then an erosion and a dilation to kill sensor noise, then the surviving pixels are connected into regions and grouped into events by time. No model is involved, and this is what produces the timestamps, durations and motion area figures.
Stage two, the model
For each event, the two or three strongest keyframes are cropped and sent for classification. What comes back is a class, a one line description and a confidence value. The video itself is not sent, only the crops.
Splitting it this way is why the numbers stay honest. Timing and area are measured, not guessed, and if the model call fails the event survives as unclassified with its measurements intact.
What it will not do
- No face recognition, no face templates, no re identification of a person between cameras or between days.
- No gait analysis, no emotion inference, no behaviour scoring.
- No licence plate reading.
- No guessing. An event the model could not classify is labelled unclassified, never assigned a plausible class.
Three people care about this, and they care about different sentences.
The person holding the phone at night
The alert channel becomes readable. Four events across a night instead of forty, and each one already says what class it was, so you know before you open the clip whether it needs you.
The operations manager
Incident review stops being an hour of scrubbing. You open the timeline, find 01:47, and the whole sequence from arrival to exit is six rows. Zones and duration are set once per camera and then left alone.
Whoever signs the invoice
A published per camera price, no hardware line, no multi year term, and a defensible answer about biometrics. On fifteen cameras that is $149 a month, or $119 billed annually, against an enterprise anchor around ten times that.
Four steps, and the first one takes about a minute.
- Try it on a clip. Export sixty seconds from your noisiest camera at its noisiest hour and run it through the demo. Detection happens in your browser and the clip does not leave your machine.
- Draw the zone you would draw. Put a polygon over the road, the footpath or the neighbouring unit, and watch the dropped counter move.
- Decide which classes matter. Most sites alert on person at all hours and on vehicle only outside opening hours, and leave animal in the timeline.
- Then connect a camera. Confirm an email address and we send you the steps for your first stream. Connecting a live camera is the part we do by email today, and this page is not going to pretend otherwise.
What a twelve camera site pays
Forecourt, entrance, two aisles, stockroom, yard, and the rest
Every tier and every line item is on the video analytics software price page. There is no free plan and no card is needed to run the demo.
What people ask before they try it.
01
Does the video leave my site?
In the demo, no: decoding and detection happen in your browser and the clip never uploads. Only small cropped keyframes of detected events go out for classification, and that is stated on the demo itself while it runs.
02
Will it run on old cameras?
Yes, if they publish a stream. Image quality matters more than camera age: a well placed 2 megapixel camera with a clean lens outperforms a badly placed 4K one, and infrared placement matters more than either.
03
How is this different from the AI in my new NVR?
The on device filter runs on a small processor with a small model, and it is genuinely useful in good light at short range. This runs a much larger model on cropped keyframes, and adds polygon zones, minimum duration and per class alerting on top. Those extras remove more alerts than the model does.
04
What happens when the internet drops?
Detection is arithmetic on frames and does not need us. Classification does, so events found during an outage arrive with their motion data and are labelled unclassified rather than being lost or guessed.
Related pages for existing cameras
RTSP software
Stream URLs per brand, sub streams and codecs.
CCTV video analytics software
The same job on an older ONVIF or DVR based estate.
Motion detection software
Why classification removes alerts that thresholds cannot.
Artificial intelligence for video surveillance
What the model does, refuses to do, and where it is weak.
Enterprise security camera systems
SSO, roles, retention, DPA and the biometrics answer.
Best video analytics software
Eight products compared, including where we lose.
Keep the cameras. Replace their brain.
Run the demo on your own footage first. When you want it on the live stream, confirm an email address and we send you the steps.