The question nobody asks first.
Before you compare detection quality, ask a simpler question: how does the AI get the video? The answer decides how long deployment takes, who has to sign off, and whether the system can follow you to your next client site.
VMS-integrated analytics connect to the video management system or the cameras directly. They use the VMS vendor’s plugin framework, an analytics server that pulls RTSP streams with camera credentials, or intelligence built into the cameras and surfaced through the VMS. If you need a refresher on the platform itself, read what a VMS is.
Screen-based analytics work from the rendered monitoring screen: the video wall, VMS client or camera dashboard the guard is already watching. The software captures the screen, separates the camera panes, identifies each view and runs detection on what is visible. It needs no camera credentials and no vendor API, because it never talks to the cameras at all.
Side-by-side comparison.
| VMS-integrated analytics | Screen-based analytics | |
|---|---|---|
| How it gets video | Camera streams or a VMS plugin API | The monitoring screen the guard watches |
| Credentials needed | Camera or VMS accounts, network access to streams | None for cameras or VMS; screen capture permission on the station |
| Works across different VMS brands | Only where a connector exists | Yes, if it renders on screen |
| Deployment | IT project: network, licensing, vendor support, change control | Install on the monitoring station and point it at the screen |
| Image quality | Full-resolution streams | Whatever resolution the pane is drawn at |
| Coverage | Every connected camera, all the time | Cameras visible on screen; cycled-off views become explicit coverage gaps |
| Client sign-off | Often needs the site owner’s IT team | Runs on the guard company’s own station |
| Vendor lock-in | Tied to the platform and its license tiers | Moves with you when the VMS changes |
What integration buys you.
Integration is the right answer in some places. Full-resolution streams give detectors more pixels to work with, which matters for small, distant objects on wide perimeter cameras. The analytics see every camera around the clock, not just the ones on screen. And a single enterprise that owns its cameras, its VMS and its IT team can justify the project once and live with it for years.
The cost is everything around the model. Integration means credentials, firewall rules, a supported VMS version, a license tier, a vendor who answers the phone and a client IT team willing to approve all of it. For a guard company or a remote video monitoring center serving dozens of clients on dozens of systems, that project repeats at every site, and some clients will simply say no.
What screen-based buys you.
Screen-based analytics trade some image fidelity for reach. If a guard can see the camera, the software can work with it, whatever brand drew the pane. There is nothing to connect, so there is nothing for the client’s IT team to approve beyond the station you already run. The same workflow, inbox and evidence trail sit on top of every site.
The honest limits are real and manageable. A camera cycled off the screen is not being observed, so good screen-based software must say so rather than assume the view is quiet. Tiny panes on a crowded wall carry fewer pixels, so layouts matter. Both are tuned during a pilot: which layouts the station runs, how panes are sized, and which views stay pinned.
When to choose each.
Choose VMS-integrated when…
- You own the cameras, the VMS and the network, and have IT staff to run it.
- Every camera must be analyzed around the clock, on screen or not.
- Your platform already ships the analytics you need at a tier you pay for.
- You need full-resolution detail on long-range perimeter views.
Choose screen-based when…
- You monitor many clients on many different camera systems.
- You cannot get camera credentials or VMS access from the site owner.
- You want AI help this month, not after an integration project.
- Your guards already watch the screen and you want to make that attention count.
Where Semaphor fits.
Semaphor is screen-based by design. It works with the camera software you already use by capturing the monitoring screen, separating each pane and giving every view a visual identity built from visual hashes, scene landmarks and a median background. Camera view tracking follows each camera as layouts cycle, resize or rearrange, and a view that drops off screen is marked unobserved, never assumed quiet.
On top of that, Semaphor runs local person and vehicle detection, publishes an alert first and then adds an AI scene report beside the original evidence. No camera replacement, no new VMS, no vendor integration project. The guard stays in control.
If your guard can see it on the monitoring screen, Semaphor can work with it.
See it on the layouts your team actually runs. Book a live demo on your own camera screens, or start a pilot on one monitoring station and confirm capture quality before you expand.
Frequently asked questions.
Does screen-based analytics need access to my cameras?
No. Semaphor never connects to cameras or the VMS. It works from the rendered screen, so there are no camera credentials to share. See security and privacy.
What happens to cameras that are not on screen?
They are marked unobserved and shown as a coverage gap. Semaphor never treats an off-screen camera as quiet.
Can I use both approaches?
Yes. Some teams keep platform analytics on owned sites and use Semaphor across client sites where integration is not an option.