Video analytics
Video analytics is software that interprets camera streams automatically, turning frames into events such as a person entering a zone, a vehicle stopping, or a crowd forming. It replaces continuous human watching with an alert on a defined condition, and it runs on the same footage a VMS already records.
What events can video analytics detect?
- Intrusion: a person or vehicle crossing a defined line or entering a zone.
- Loitering: presence in an area for longer than a set duration.
- Object left behind or removed, relative to a learned background.
- Crowd density and queue length, counted rather than estimated by an operator.
- Vehicle events: wrong-way travel, stopped vehicle, and licence plate reading where permitted.
- Personal protective equipment checks, such as a missing helmet or vest in an industrial area.
Each of these is a rule over detections rather than a separate technology. The system detects and tracks objects, and an event is a condition expressed over those tracks: where they went, how long they stayed, what class they were. Understanding that structure is what makes it possible to predict which rules a site can support.
How is video analytics different from motion detection?
| Motion detection | Video analytics | Face recognition | |
|---|---|---|---|
| What it reports | Pixels changed | A person entered zone 3 | This is an enrolled person |
| Distinguishes object class | No | Yes | Yes |
| Triggers on rain, foliage, headlights | Frequently | Rarely | Rarely |
| Identifies individuals | No | No | Yes |
| Typical regulatory burden | Low | Low to moderate | High |
Motion detection reports that pixels changed, which is why weather and headlights set it off. Video analytics classifies what moved and applies a rule to it. Face recognition goes further and identifies who, which is a different legal category: analytics that counts people is not doing biometrics, and the distinction matters in a privacy assessment.
What governs the false alarm rate?
- Camera placement: an angle that shows a person's full height produces far better tracking than a ceiling-mounted overview.
- Resolution at the point of interest, not at the centre of the frame, since detection fails at the edges first.
- Lighting, including headlights at night and direct sun into the lens at particular hours.
- Rule tuning: a dwell time or minimum object size removes most nuisance alerts without weakening the rule.
- Scene change management, because a moved camera or a new sign invalidates the zones drawn against the old view.
Most disappointing analytics deployments are placement problems rather than model problems. A site survey before procurement predicts the outcome better than any comparison of detection claims.
Frequently asked questions
- What is video analytics?
- Video analytics is software that interprets camera streams automatically, turning frames into events such as a person entering a zone, a vehicle stopping or a crowd forming. It replaces continuous human watching with an alert on a defined condition, running on the footage a VMS already records.
- What is the difference between video analytics and motion detection?
- Motion detection reports that pixels changed, which is why rain, foliage and headlights trigger it. Video analytics classifies what moved, tracks it, and applies a rule such as entering a zone or staying too long. The classification step is what removes most nuisance alarms.
- Is video analytics the same as face recognition?
- No. Analytics detects and classifies objects and applies rules to their movement without identifying anyone. Face recognition determines who a person is by matching against enrolled templates. Counting people is not biometric processing; identifying them is, and the two sit in different regulatory categories.
- Does video analytics work on existing cameras?
- Usually, if they deliver standard ONVIF or RTSP streams and the view suits the rule. Placement matters more than the camera model: a ceiling-mounted overview tracks people poorly regardless of resolution, and rules drawn against a badly placed view produce false alerts no tuning fixes.
- Why does video analytics produce false alarms?
- Most often because of placement, lighting or an untuned rule rather than the detection model. A dwell time, a minimum object size and a zone drawn to exclude a road usually remove the bulk of nuisance alerts, and a moved camera invalidates the zones drawn against the old view.
- Can video analytics run on-premise?
- Yes, and for multi-camera sites it usually should, because sending every stream to a remote service is a continuous upload commitment. On-premise appliances analyse the streams locally and forward only events and clips, which also keeps footage inside the building.