AI video analytics software features
Grouped by the job it does rather than by the module it lives in.
The last group is the one nobody publishes: what this does not do yet.
- Five classes, no identities
- Any ONVIF or RTSP camera
- Published pricing, no call
01
Whatever is on the wall, without an installer.
Getting video in
- ONVIF discovery
- Point it at the subnet and it finds the cameras that answer, with their profiles and stream URLs.
- RTSP by URL
- Paste the stream address if discovery is off or the camera is on another network. This is the path that works on the oldest hardware.
- Sub stream aware
- It reads the low resolution copy your camera is already producing, so the main recording stream is untouched and your recorder keeps its bandwidth.
- H.264 and H.265
- Both, including the H.265 profiles that older NVR software refuses.
- Uploaded clips
- MP4 and WebM, up to 60 seconds in the demo. Useful for testing a scene before you point anything at it.
- Cameras it works with
- Hikvision, Dahua, Reolink, Axis, Amcrest, UniFi, Lorex, Vivotek, Hanwha and anything else that speaks the two protocols above.
02
The detection stage, with every knob that matters exposed.
Finding what moved
- Adaptive background model
- The empty scene is learned and re-learned as daylight drifts, so dusk does not read as motion across the whole frame.
- Sensitivity
- One control, 1 to 10, driving the luma delta threshold and the minimum region area together. It genuinely changes the numbers, which you can watch happen in the demo.
- Noise suppression
- Erosion and dilation before anything is grouped, so a single hot pixel and an infrared moth do not become an event.
- Event grouping
- Detections in neighbouring frames are joined into one event with a start, an end and a duration, instead of ninety separate triggers for one person walking.
- Motion area
- Every event carries the percentage of the frame that moved, which is the fastest way to tell a person from weather without opening the clip.
- Minimum duration
- From 0.2 to 5 seconds. This one control removes most single frame flicker on its own.
03
The stage your recorder never had.
Deciding what it was
- Five classes
- person, vehicle, animal, package, other motion. Custom classes from the Operate tier up.
- Confidence on every event
- A real value, shown as a percentage, so you can see when the model was unsure rather than being told everything is certain.
- One line description
- What is visibly happening in the crop, written out, so the event list reads without opening footage.
- Honest failures
- If classification is unavailable the event stays in the timeline marked unclassified with its motion data intact. No guessed label, ever.
- No biometrics
- No face templates, no face matching, no gait, no re-identification. A product decision, not a setting.
04
The filters between an event and your phone.
Deciding what you hear about
- Exclude zones
- Click a polygon over the road, the tree or the neighbour's window. Events whose centroid falls inside are dropped. Two zones per camera on Watch, unlimited above it.
- Class filters
- Alert on people at night and on vehicles only during opening hours, or on nothing at all and just keep the timeline.
- Schedules and escalation
- Different rules for different hours, and a second recipient if the first does not acknowledge. Multi-Site and up.
- Email, webhook, Slack
- Email on every plan. Webhook and Slack from Operate. SIEM and syslog delivery on Enterprise.
- The timeline keeps everything
- Filtered events are not deleted, they are just not an alert. You can still search them later, which is what makes a filter safe to set aggressively.
05
The part the person signing the invoice reads.
Living with it in a company
- Roles
- Viewer and Operator from Operate, plus Admin and custom roles from Multi-Site.
- Per site scoping
- A contractor sees one yard, not the estate.
- SSO
- SAML and OIDC from Multi-Site up.
- Audit log
- 30 days on Operate, a year on Multi-Site, custom on Enterprise. Who viewed which clip, who changed which zone, who exported what.
- Retention
- A stated window with configurable deletion, and a DPA available from Operate.
- Procurement
- Annual invoicing, purchase orders, a W-9, and a security questionnaire answered rather than deflected.
The mechanism behind groups two and three is written out stage by stage on artificial intelligence video surveillance, and group five is expanded on enterprise security camera systems.
06
What it does not do
Two things a competitor has that we do not, written down here rather than discovered later.
Camera health and PTZ control
It reads the stream and classifies what is in it. It does not tell you a camera went offline, and it does not drive a pan, tilt and zoom head. If that is the job, an enterprise platform or your existing NVR still does it and this sits alongside.
A hardware warranty
We sell software for cameras you own, so if the camera fails, that is between you and whoever sold it. An enterprise vendor replacing the whole system carries that risk for you, and charges for it.
07
Read next
The same capabilities, from a different angle
Person detection camera software
The one class most buyers arrive asking about.
Camera analytics
What an event timeline replaces, and how long review takes after it.
CCTV video analytics software
The same features against an older ONVIF install.
Video analytics software price
Which of the five groups above sits in which tier.
Video analytics platform use cases
The features arranged by industry instead of by job.
Intelligent video analytics software FAQ
Accuracy, privacy law, cost and the questions under them.
Every feature above starts with one email address.
Confirm it with a 6 digit code, pick the tier that matches your camera count and pay by card. The demo stays open either way.