What CCTV analytics actually is
CCTV analytics applies deep-learning classifiers and rule logic to camera feeds in order to generate actionable detection events rather than continuous recording. Modern platforms classify people, vehicles and sub-categories, then apply spatial and behavioural rules such as line crossing, loitering, direction of travel and object removed.
The output is a stream of qualified alarms an operator can act on — not raw motion pixels an operator has to interpret.
Edge, server and hybrid deployment
Edge analytics runs the classifier on the camera itself. This minimises bandwidth, reduces latency and keeps detection working even if the network to a central VMS is disrupted. Server analytics centralises processing on a GPU-backed appliance or cloud tenant, allowing heavier models and multi-stream fusion.
Most large deployments are hybrid: fast per-camera classification at the edge, deeper cross-camera analytics centrally for tracking and forensic search.
- Edge — low bandwidth, resilient, tightly scoped models
- Server — heavier models, cross-camera correlation, forensic search
- Hybrid — edge triage plus server verification and long-term analytics
Which cameras support analytics
Analytics-capable cameras carry a defined processor (often an NPU or dedicated analytics SoC) and provide sufficient resolution for the intended detection distance. Retrofitting analytics to older cameras through a server layer is possible but degrades detection quality — codec artefacts, low frame rate and poor low-light performance all reduce classifier accuracy.
The most cost-effective analytics upgrade for a mixed estate is usually to replace the small number of true detection cameras with current-generation units and leave overview cameras on their existing hardware.
Tuning and commissioning workflow
A structured tuning phase in the first thirty days after commissioning defines the difference between analytics that operators trust and analytics they disable. Tuning covers detection zone geometry, sensitivity thresholds, classification categories, scheduled rules and privacy masking.
The workflow should be evidenced: recorded before-and-after alarm volumes, false-positive categorisation and operator acceptance testing. Without evidence, tuning is anecdotal and drifts.
Measuring analytics performance
Meaningful performance metrics are: verified alarms per week, false-alarm ratio, missed-event rate against a documented test walk, mean time to operator verification, and detection accuracy at defined ranges. Vendor confidence scores in isolation are not a substitute.
Contracts should include a right to run scheduled test walks and a defined remediation SLA when metrics fall below target.
Frequently asked questions
Is CCTV analytics the same as motion detection?
No. Motion detection triggers on any pixel change, including wildlife, weather and shadows. CCTV analytics classifies objects and applies rules, so it can distinguish a person crossing a line from leaves in wind. The typical false-alarm reduction from motion to analytics on the same camera exceeds ninety percent when the analytics is correctly tuned to the scene.
Can analytics run on my existing cameras?
Sometimes, via a server-side analytics layer, but performance depends on camera resolution, codec and low-light capability. Retrofitting older cameras rarely matches current-generation analytics hardware. The pragmatic upgrade path is to replace the specific cameras that need to detect intrusion with analytics-capable units and leave overview cameras alone.
How long does analytics tuning take on a new site?
Expect around thirty days of iterative tuning covering zone geometry, sensitivity, classification categories and scheduled rules. This phase is where a well-specified system either becomes trustworthy or becomes shelfware. Skipping the tuning phase is the single most common cause of monitored analytics being switched off within twelve months of commissioning.
Does analytics work at night?
Modern analytics performs well at night when combined with adequate infrared illumination or dual-spectrum thermal cameras. Classifier accuracy in true darkness on a standard visible-light camera without IR support falls off sharply. Camera specification and lighting design are analytics decisions, not separate ones — they must be planned together at the design stage.
How is analytics performance measured?
Useful metrics are verified alarms per week, false-alarm ratio, missed-event rate against a documented test walk, and mean time to operator verification. Vendor confidence scores in isolation are marketing rather than performance data. Contracts should give the site owner a right to run scheduled test walks and enforce a remediation SLA where measured metrics fall below the design targets.
Can analytics detect vehicles and number plates?
Yes — vehicle classification is standard, and dedicated automatic number plate recognition (ANPR) analytics is widely available. ANPR uses specifically configured cameras and lighting angled for plate visibility, so it is usually deployed on gates and access lanes rather than general perimeter cameras. Combining ANPR with intruder analytics is common on logistics and industrial sites for access control integration.
Does analytics create privacy risks?
Analytics that classifies objects generically — person or vehicle — does not by itself identify individuals and sits well within UK GDPR and US state privacy frameworks. Analytics that identifies individuals (facial recognition, gait) has significantly higher compliance obligations, including a lawful basis, DPIA and often signage. The two categories should not be treated as one class of technology for compliance purposes.
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 cctv analytics
Tell us about your site and we'll connect you with a commercial security specialist who understands your detection, monitoring and response requirements.