This glossary is written for people who work the desk, run the monitoring center or buy the service. It covers the monitoring operation, the AI underneath modern alerts and the evidence that backs up a decision. For how these pieces come together in one product, read what Semaphor is; for the fundamentals, browse the guides.
Monitoring operations.
- GSOC
- A global security operations center: a central room where a corporate security team monitors cameras, alarms and incidents across many facilities. See what is a GSOC.
- SOC
- A security operations center. In physical security it means a monitoring room for cameras and alarms; in IT it usually means a cybersecurity team watching networks and systems.
- Video wall
- A grid of displays or one large screen showing many camera views at once, typically in a monitoring center or GSOC. It shows everything and prioritizes nothing on its own.
- Remote video monitoring
- Watching a site’s cameras live from an off-site monitoring center rather than from a desk on the property. See what is remote video monitoring.
- Virtual guarding
- A guarding service delivered through cameras: remote operators watch live views, challenge intruders through speakers and dispatch police or patrol. See what is virtual guarding.
- Remote guarding
- Another name for virtual guarding, often used for proactive live monitoring with audio talk-down. Compare it with a physical post in virtual guarding vs on-site guards.
- Post orders
- The written instructions for a guard post: what to watch, what counts as an incident and who to call. For camera work, see post orders for camera monitoring.
- Alarm verification
- Confirming with video or audio that an alarm reflects a real event before requesting a police response. See alarm verification for police response.
- False alarm
- An alarm or alert that fires when no real security event occurred, from a faulty sensor, user error or an environmental trigger. Repeated false alarms can cost a site its police response priority.
- Nuisance alarm
- An alert caused by something real but irrelevant, such as rain, headlights, shadows, animals or swaying trees. See how to reduce nuisance alerts.
- Alert triage
- Sorting incoming alerts by urgency so the most important ones are reviewed first. See security alert triage and priority levels.
- Coverage gap
- A period when a camera is not being observed, for example because a layout cycled it off screen. Semaphor marks these views unobserved rather than assuming they are quiet; see camera view tracking.
- Tailgating
- An unauthorized person following an authorized one through an access-controlled door before it closes. See tailgating at access doors.
- Loitering
- A person lingering in an area without an apparent purpose, often a precursor to theft or trespass. See loitering detection.
Cameras and video systems.
- VMS
- A video management system: software that records, organizes and displays camera feeds from many cameras. See what is a VMS.
- NVR
- A network video recorder: a device that records video from IP cameras over the network, often with a built-in viewing interface.
- DVR
- A digital video recorder: a device that records video from analog cameras, digitizing the signal on the recorder rather than in the camera.
- PTZ
- A pan-tilt-zoom camera that can be steered and zoomed remotely or by preset. See PTZ camera tours.
- Camera tour
- A programmed sequence that moves a PTZ camera between preset positions, or cycles a monitor through a set of camera views. Either way, views not currently shown go unwatched.
- AI video monitoring
- Live camera monitoring assisted by AI that watches continuously and surfaces the moments that need a person. See what is AI video monitoring.
- Video analytics
- Software that analyzes video automatically to detect objects, events or patterns, also called video content analysis. See what is video analytics.
- Edge analytics
- Video analytics that run on site, in a camera, appliance or workstation, instead of in a remote data center. See edge vs cloud video analytics.
AI and detection.
- Object detection
- A model that finds and labels objects in an image, such as people and vehicles, and reports where they are. See how object detection works on security cameras.
- Bounding box
- The rectangle a detector draws around an object it has found, marking its position and size in the frame.
- Confidence score
- A number a detector attaches to each detection to express how sure it is. It is a model’s estimate, not a certainty, and systems set thresholds on it.
- Tracking
- Linking detections of the same object across frames so one person walking through a scene is one continuous track rather than many separate hits. See person and vehicle detection.
- Track ID
- The number a tracker assigns to one object across frames. In Semaphor, track IDs are local to a camera clip and do not identify who a person is.
- Occlusion
- When an object is partly or fully hidden behind something else, like a person passing behind a pillar. Good trackers recover the track when the object reappears.
- Vision-language model
- An AI model that takes images and text together and can describe a scene in words. See vision-language models for security cameras.
- Scene report
- A short written assessment of a camera event: what is happening, what was observed, a recommendation and what remains uncertain. See AI scene reports.
- Human-in-the-loop
- A design where AI prepares information but a person makes the decision. See human-in-the-loop security AI.
- Preliminary alert
- An alert published immediately when activity is detected, before the AI review finishes, so the guard never waits to see that something moved. See alert-first review.
- Duplicate suppression
- Grouping or holding back repeat alerts about the same activity, often with cooldowns, so one parked car does not alert all night. See duplicate alert suppression.
Evidence and security.
- Evidence hash (SHA-256)
- A cryptographic fingerprint of an exact file. If a single pixel changes, the hash changes, which shows the evidence has not been altered. See evidence and audit trail.
- Perceptual hash
- A fingerprint of what an image looks like rather than its exact bytes, such as pHash or dHash. Similar-looking images produce similar hashes, which helps match camera views and near-duplicate frames.
- Chain of custody
- The documented record of who captured, handled and transferred evidence and when. See incident report video evidence.
- Mutual TLS
- An encrypted connection where both client and server prove their identity with certificates. Semaphor supports it for hosted AI review; see security and privacy.
Missing a term? Tell us, or start a pilot and hear the vocabulary in context on your own sites.