Video management system

Video management system
Before
Cameras from several makers, each watched through its own software, with no way to see every site at once.
Now
Every camera on every site in one place: recorded, searchable, and watched by AI that raises an alarm when something matters.
The work
It began with a partner who runs a camera security business and knew exactly where the existing software fell short. Each maker's cameras were watched through that maker's own system, sites were never tied together, and the AI on offer was locked to one vendor's hardware.
We built one video management platform that takes standard IP cameras from any maker, records every site, and puts live and recorded video in front of the people who need it: in a browser, on a video wall or on a phone. AI watches the streams and raises an alert only when something matters, and a person can search across every camera and export the clip they need.
Underneath, it is four pipelines working together: one that brings each stream in, one that records it, one that runs AI on it, and one that turns what the AI finds into alarms a person acts on. Each can be changed or scaled without touching the others, and the AI models can be swapped to suit the hardware at each site.
Cameras it works with
Standard IP cameras, whoever made them.
- Axis
- Avigilon
- Sony
- Any ONVIF camera
Cameras are added through ONVIF, the industry standard most IP cameras speak, and streamed over RTSP. So a site can mix makers, keep the cameras it already owns, and choose the next ones on price and quality rather than on which software they are tied to.
How it works
From the camera to the person watching, in six stages.
- 01ConnectEach camera is found and added through ONVIF, its streams are pulled over RTSP, and its credentials are stored encrypted, never in plain text.
- 02IngestA media server takes each stream once and shares it out to recording, live view and the AI, so no camera is asked for the same video three times.
- 03RecordThe recording pipeline, the system's NVR, writes every stream in short segments to storage, with how long it is kept set for each site.
- 04AnalyseThe AI pipeline samples frames and runs detection models on them, for people, vehicles and more. Models are chosen to suit each site's hardware, and can be swapped or run on the client's own equipment.
- 05AlertWhat the AI finds becomes an event. Events pass through each site's rules into the alarm manager and out as notifications, and a person acknowledges every alarm.
- 06Watch and searchLive and recorded video play in the browser with low delay, on a single screen or a video wall. Search runs across cameras and time, and clips export into case files.
Under the hood
For the technical reader: what each stage is built with.
| Stage | What it does | Built with |
|---|---|---|
| Connect | Camera discovery, streams, credentials | ONVIF, RTSP, encrypted credential store |
| Ingest | One pull per stream, shared to every consumer | MediaMTX media server |
| Record (NVR) | Segmented recording, retention per site | Object storage, per-site retention rules |
| Analyse (AI) | Detection on sampled frames, per-site models | YOLO-family models, hardware-matched, swappable |
| Alert | Events, rules, alarms, notifications | Kafka event stream, rules engine, alarm manager |
| Watch and search | Live view, playback, video wall, clip export | WebRTC and HLS in the browser, React front end |
| Platform | Sites, users, roles, audit | Django back end, domain-driven design, role-based access |
What comes next
- AI pipelineAn agent that tunes the AIAn agent that watches how well detection is working at each site, and recommends better models or settings for its cameras and hardware. It learns site by site, and nothing is pooled between customers.
- EdgeAI beside the camerasDetection running on edge computers at the site, so only events travel, not every frame. It keeps working when the connection does not.
- ModelsModels trained for the workCustom open models trained on each customer's own scenes, with their agreement, so detection gets sharper where it matters most to them.
Built to be trusted
- Each site is its ownFootage, settings, rules and models are kept apart for every customer and every site.
- Access by roleOperators, supervisors and administrators see and do only what their role allows, with two-factor sign-in.
- Every action recordedWho watched, exported or changed what, and when, is kept on the record.
- No vendor lock-inAny ONVIF camera, swappable AI models, and the code and infrastructure owned by the client.
Built for
A security integrator, and the sites it protects
Security · In production
What it does
- IP cameras from any maker, on one platform
- Live view in the browser, on a video wall or on a phone
- Continuous recording, with retention set for each site
- AI detection, with alerts and an alarm manager
- Search across cameras, clip export and case files
- Each site's footage, settings and models kept its own
Want to see it running?
Where the client agrees, we will walk you through the live system on a call.