Object classification
Modern analytics classifies objects in the scene — person, vehicle, sometimes more granular categories. The classifier runs on the camera (edge analytics) or on a server, depending on deployment.
Classification is what allows the system to ignore wildlife, weather and routine movement and only escalate events involving people or vehicles.
Analytic rules: lines, zones, loitering
Detection rules are applied on top of classification — line crossing, zone intrusion, loitering, direction of travel, object left or removed. The right rule combination reflects how an intruder would actually behave on the site.
- Line crossing — a virtual tripwire at the boundary
- Zone intrusion — a defined area is entered
- Loitering — a person remains too long in an area
- Direction of travel — only one direction triggers detection
False-alarm reduction
Well-tuned analytics dramatically reduces false alarms compared with basic motion detection. The gain comes from classification (ignoring non-target objects), tight detection zones and rule logic that mirrors real intrusion patterns.
What video analytics still struggles with
Heavy rain, fog, lens fouling, low contrast scenes and unusual camera angles all degrade analytics performance. So do scenes dominated by vehicle movement near the detection line — direction and class filters help but don't eliminate the issue.
Analytics is a powerful detection layer, not a complete solution. It earns its place inside a wider design that includes monitored response and (often) other sensor types.
Edge vs server analytics
Edge analytics run on the camera itself, reducing bandwidth and latency. Server analytics centralise compute and allow more complex models. Modern designs typically blend the two — basic classification at the edge, deeper analytics centrally.
Frequently asked questions
Is video analytics the same as AI?
Modern video analytics typically uses deep learning models — which are commonly badged as AI. The relevant question is performance: classification accuracy, false-alarm rate and behaviour in adverse conditions, not the marketing label.
How accurate is human/vehicle classification in practice?
On well-installed modern systems, classification accuracy in normal conditions is high — typically well above 95 percent in independent testing. Performance falls off in extreme weather, low light without IR support, and at long range.
Can analytics be added to existing CCTV cameras?
Sometimes via a server-side analytics layer, depending on camera resolution and codec. Often the better answer is to replace key detection cameras with current-generation analytics-capable units rather than retrofit older hardware.
Which sites benefit most from analytics?
Any site where the camera layer is intended to detect intrusion — not just record it — benefits from analytics. This includes perimeters, yards, remote infrastructure, out-of-hours retail and vacant premises. Sites where cameras are used purely for post-event evidence retrieval derive less operational value from analytics, though basic classification still improves footage searchability significantly.
How do analytics platforms handle privacy masking?
Modern platforms support configurable privacy masks that blank out non-relevant zones — public footpaths, neighbouring property, staff break areas — while allowing analytics to run on adjacent detection zones. Masks are enforced at the camera or platform level and are part of the compliance record maintained under GDPR in the UK and comparable state-level privacy law in the US.
Can analytics detect specific behaviours like fighting or falls?
Some platforms offer behaviour analytics — aggression, falls, unusual crowd density — but performance varies substantially by scene and use case. Behavioural detection is more mature in specific contexts like healthcare fall detection than in generalised aggression detection, which remains challenging. Deployment decisions should follow validated evidence rather than vendor demonstration footage across the market currently.
How is analytics tuned after installation?
Tuning is an iterative process during the first thirty days: zones are refined, sensitivity thresholds adjusted, and classification categories confirmed against real site activity. A structured tuning phase is what separates production-grade analytics from a persistent false-alarm problem. Skipping this phase is the single most common cause of monitored CCTV systems being switched off within their first year.
Continue in the intruder detection hub
The intruder detection hub sets out how this technology fits alongside the other layers of a complete commercial design.
Speak to a specialist about video analytics
Tell us about your site and we'll connect you with a commercial security specialist who understands your detection, monitoring and response requirements.