Overview
Aegis Vision watches a site’s existing camera feeds continuously and raises an alert on the site’s own guard console when it detects a weapon or a safety incident. All processing happens on a server inside the building.
What it detects
Section titled “What it detects”| Category | Examples | How |
|---|---|---|
| Weapons | Handgun, rifle, knife or blade, explosive device, blunt melee weapon | Object detection on each frame |
| Safety incidents | Fights and assaults; a person who collapses and stays down; an unattended bag; a guard post left empty | Body-pose sequences, object dwell time, zone presence |
The system does not identify people. It does not match faces against any database. The optional guard-post module recognises only the enrolled guard’s face, on the camera covering that post, and only when a site enables it.
How detection works
Section titled “How detection works”- Detect. A fast detector scans every frame from every camera for candidate objects and body poses.
- Verify. A larger model confirms the detection. Borderline detections must persist across frames and be corroborated, or they are dropped. The alert is raised at this stage.
- Classify. Two independent classifier models grade the incident type and confidence after the alert is already on the console.
- Respond. A guard confirms, dismisses or escalates from the console. Escalation to phones and emergency contacts follows the site’s configuration.
Where data goes
Section titled “Where data goes”- Live video is read from the cameras or recorder and processed in memory on the site server.
- Alert images (the frame that triggered the alert) are stored on the site server.
- The guard console runs on the site’s local network.
- The Aegis Vision operator reaches the server over an encrypted private network for support and model updates, and can view alerts a guard has already resolved. Nothing is cached outside the country.
- Only when a school administrator presses Send to Hub are cropped, face-blacked-out, metadata-stripped detection crops transferred for model improvement.
Retention of alert images is set per site and recorded in the deployment agreement.
Aegis Vision · Trust Centre · Privacy · Legal notice · Report a vulnerability