Observe what’s visible.
Work from the camera views on a monitoring screen. Follow visible feeds as layouts change, with gaps made explicit.
Works with your existing camera software →
SECURITY INTELLIGENCE. HUMAN
JUDGMENT.
More cameras.
Clearer priorities.
A second set of eyes for your security operation.
Semaphor is AI video monitoring for security operators: it turns the camera views your team already watches into focused alerts, supporting evidence, and context to act. No camera replacement, no new VMS.
01 / SEE THE WORKFLOW
From activity to understanding.
Continuous research footage. Synchronized detection overlays.
See the scene’s AI report alongside the original footage.
SEMAPHOREvent workspace
RECORDED DEMO
Stairwell / G419
Video could not load. Select another camera or reload the page.
01 PersonV01 VehicleClick a number to inspect its track.
TRACK INSPECTOR
Waiting for video
Select a track
- Detector score
- —
- Track span
- —
- First detected
- —
- Last detected
- —
- Image X / Y %
- —
Waiting for video.
Scores are detector outputs, not accuracy percentages. Span includes detection gaps. Trails follow box centers over the past 1.5 seconds; IDs are local to this clip.
Original evidence stays beside the assessment.
Continuous prerecorded video with detections synchronized to playback. The AI reviews are saved results, not live inference. These ordinary-activity examples do not establish incident accuracy. Numbered labels follow estimated tracks within each clip. Overlapping vehicle duplicates are merged. Occlusion can still cause missed detections or ID switches; these are not verified identities. Want to see it on your own monitoring screens? Book a live demo.
About the footage & results
Source: MEVA KF1 dataset, Kitware / IARPA DIVA; Corona, Osterdahl, Collins and Hoogs (WACV 2021). Used under CC BY 4.0. Clips are shortened, resized and re-encoded without audio; detections are a separate overlay, measured on those clips. AI assessments were recorded from sampled stills of the same interval and can be wrong. Review controls illustrate the workflow; their timing is not a speed benchmark. View the source manifest and recorded outputs.
02 / A LAYER OF ATTENTION
Your cameras. Your people.
A more focused workflow.
Designed around the monitoring screen your operators use, with the guard at the center of every decision. See how it works →
Bring activity forward.
Use people, vehicle and scene-change detections to gather relevant evidence and surface an initial alert.
Person and vehicle detection →Add context. Keep control.
Review an AI assessment alongside the original evidence. The guard decides what needs action and what happens next.
AI scene reports →03 / BUILT FOR SECURITY OPERATORS
Augment the operation
you already run.
For security guard companies, remote monitoring centers and security operations centers that need their guards to cover camera-heavy sites with clearer context and less manual searching.
Talk to salesStart with one monitoring station.
Evaluate capture quality, dashboard layouts and the situations your operators actually need to review. How a pilot works →
Measure the operational difference.
Look at missed events, nuisance alerts, response delay and time spent reviewing evidence.
Expand on evidence.
Use pilot results to decide where assisted monitoring earns a place in your operation. See use cases by site type →
TALK TO SALES
Let’s talk about
your operation.
Leave your email address or phone number. Our sales team will reach out to book a live demo on your own camera screens.
Does it work with our existing camera software?
Yes, that’s the point. Semaphor observes the rendered camera views on your monitoring screen instead of depending on a specific vendor’s integration, so you keep the VMS you already use. We confirm compatibility on your screens, layouts and image quality during a pilot. Only footage visible to the capture is observed; a camera cycled off-screen creates a coverage gap. How it works with existing software →
What can we evaluate today?
Camera-pane separation, visual camera matching, person and vehicle detection and an alert-first review loop. This public demo replays selected detector results and saved AI assessments; a live demo and a pilot measure detection, tracking and guard workload on your own sites.
Does the AI decide whether to dispatch?
No. The guard remains responsible for action. AI assessments are supporting information, shown with the original evidence and uncertainty. Ordinary people or vehicle activity does not by itself establish a security incident. Why human-in-the-loop matters →
Is any customer footage shown or collected here?
No customer footage is shown or collected here. This website uses public research footage with attribution and does not request access to your screen, cameras or microphone. If you submit the sales form, we save the contact details you provide for follow-up for up to 90 days.