Object classification
AI classifies what's in the scene — person, vehicle, animal, bag — and only triggers when a target object is present. This is the central reason modern analytics outperform pixel-motion detection.
Common analytic rules
The rules layered on top of classification:
- Line crossing (virtual tripwire, with direction)
- Intrusion zone (object inside a defined polygon)
- Loitering (object remains in zone beyond threshold)
- Direction & speed
- Object left behind / removed
Edge, recorder or cloud
Analytics can run on the camera (edge), on a recorder (NVR) or in the cloud. Edge is the modern default — lower latency, lower bandwidth, lower cost at scale.
In summary
- Classification + rules is the modern analytics stack.
- Edge analytics is the dominant deployment model.
- Tuning is what separates production-grade analytics from noise.
Frequently asked questions
Do all modern cameras have analytics?
Most current commercial cameras include on-camera analytics. Capability varies — basic motion vs full object classification vs licence-plate-grade.
What accuracy should I expect from a modern classifier?
Independent testing of current-generation edge classifiers reports human and vehicle detection accuracy well above ninety-five percent in normal conditions. Accuracy degrades in extreme weather, at long range without sufficient pixel density on target, and where scene contrast is poor. Vendor-claimed accuracy figures should always be validated against your own site before commissioning.
Can analytics distinguish employees from intruders?
Not reliably by identity — analytics classifies object type, not individual identity. Behaviour and access-time context can distinguish routine from anomalous activity: an employee crossing a delivery yard at three in the morning is anomalous, regardless of identity. Facial recognition is a separate technology layer with materially different legal and operational considerations in both markets.
Does analytics work at night?
Yes, if the scene has adequate infrared illumination or the camera is a thermal sensor. In an unlit scene without infrared, analytics performance falls off sharply below usable pixel density on target. Night detection design is a specific discipline — throwing more cameras at the problem without addressing illumination almost never produces the intended detection outcome.
How much processing power does edge analytics need?
Modern commercial edge analytics runs on dedicated processors built into the camera and adds no observable load to the main streaming function. Older cameras retrofitted with analytics via firmware update typically deliver limited capability. When analytics is a design requirement, cameras should be selected specifically for that role rather than repurposed from an existing overview role.
Where does the analytics licence sit?
Depending on architecture, analytics licences are held on the camera itself, on the NVR or VMS, or in a cloud analytics platform. Cloud licensing is typically a recurring per-camera fee, edge licensing is typically capital. Total ten-year cost is broadly similar across models; the trade-off is between recurring cost predictability and up-front capital expenditure.
Are there privacy rules around analytics footage?
Yes — UK GDPR and US state privacy laws treat CCTV footage, including analytics-derived metadata, as personal data. Retention, access, subject-access response and lawful-basis documentation all apply. The specific compliance burden depends on jurisdiction and how the footage is used, and specialist advice is recommended before deploying analytics for anything beyond conventional security purposes.
Talk through this with a specialist for your site
Tell us about your site and we'll connect you with a commercial security specialist who understands your detection, monitoring and response requirements.