Have you read part 1 of this article?
In part 1 we go through cameras, security risks, storage and cabling.
Thinking about getting a new CCTV system?
Part 2: Integration, analytics, AI, cloud and costs
Moving to a modern cloud based service brings real benefits. Predictable subscription costs and automatic software updates, but there are some catches and drawbacks where a hybrid system may be better, as we explain below.
Kris Simons
Sales Director, ATEC
“Cameras, storage, software, analytics and cloud can be reviewed separately. Replace what is out of support, restricted by procurement, or failing its operational requirement. Keep what still works, and introduce new capability where it changes an outcome.”
In this article
Share on:
We want to integrate with other systems
Bringing CCTV together with access control and intruder detection lets operators verify alarms with video from one interface. It cuts operator effort and improves response decisions.
There are three architectures. The VMS as the front end, with access and intruder events pulled into it. The access control system at the centre, with video pulled into that. Or, on larger multi-vendor estates, a PSIM or situational awareness layer above both, which costs more and adds a platform to maintain but avoids tying your front end to one manufacturer’s roadmap.
Whichever you choose, alarms need to land somewhere clear and immediate, ideally a map showing doors, cameras and alarm points together, so the operator can see where the alarm is and pull the video without switching screens.
Most enterprise platforms have established APIs. But an API is not an integration. What matters is whether a certified, vendor-tested integration exists between your specific versions, who maintains it after either system is upgraded, and what the connector licence costs. ONVIF Profile A and Profile C are meant to help, but adoption across access control manufacturers is patchy.
Worth raising at design stage: linking video to access control creates a combined record of identified people and their movements, which is a DPIA trigger in its own right.
We want analytics and AI
Traditional analytics trigger alerts: line crossing, intrusion, motion within a defined area. AI classifies what it sees, distinguishing a person from a vehicle, identifying vehicle types, clothing colour, or whether someone is carrying a bag. That metadata makes alerts more accurate and post-incident searches considerably quicker.
Processing runs either at the edge in the camera or centrally on servers handling multiple streams. Central processing means older cameras can contribute metadata they cannot generate themselves, which is often presented as a free way to reuse an existing estate. It is not. GPU capacity scales per stream, along with decode load, power and cooling.
Accuracy on legacy cameras is also worse than a demo suggests. Low frame rates, aggressive bitrate caps, compression artefacts, poor low light and mounting heights chosen for a different purpose all degrade detection. “Good enough image quality” needs measuring against the operational requirement.
Two things get lumped together as AI and should not be. Object classification and attribute search sit in a very different legal position from facial recognition, which processes biometric data and carries a much heavier assessment burden.
Capability can be introduced gradually, starting with higher-risk areas or a central platform alongside the existing VMS. Decide first what you want it to do, whether that is fewer false alarms, faster searches or catching something you currently miss, then measure whether it does. And fix the basics. A badly positioned camera in poor light will defeat any algorithm behind it.
Should we move to the cloud?
Most manufacturers now offer a cloud or SaaS platform. For smaller systems there are few downsides. For large estates it is a harder call, and hybrid is often the answer.
The advantages are real. Subscription costs are predictable, the software stays current, and there is little or no on-premise hardware to patch, monitor and replace. Predictable is not the same as cheaper, though. Check retention tiers, data egress charges and what the per-camera price does at renewal.
Bandwidth is the constraint most often underestimated. A 4MP H.265 camera runs at roughly 2 to 4 Mbps under moderate activity, so 200 cameras is 400 to 800 Mbps of sustained upload, all day every day. Few sites have that without a dedicated circuit. If the control room needs instant access during an incident, work through what an outage or a degraded link actually does.
Hybrid avoids most of that. Record locally first, either to SD card in the camera or a small NVR, then upload after 24 to 48 hours. Bandwidth demand drops and you still get off-site retention. If you use SD cards, specify industrial-grade media and monitor card health, because they wear out and tend to fail quietly. For anything evidential, local NVR recording is safer.
Cloud also raises questions you would not otherwise ask: where the footage is physically stored, who the processor is, which sub-processors are involved, and who can access it.
What will the licensing cost be?
This is often the question underneath all the others. VMS cost is not a single number. There are per-channel licences, the annual maintenance agreement or subscription, and connector licences for any integrations. The common trap is a lapsed maintenance agreement, where reinstating after several years can approach the cost of a new licence. That is frequently what turns “we’ll upgrade eventually” into “we need to decide now”.
Compare over five years or more and include the circuit upgrade if you need one, power, cooling, rack space, the migration project, training and decommissioning. SaaS also shifts capital expenditure to operating expenditure, which suits some funding models and causes problems for others.
Ask one more question before you sign anything. How do we get our footage out if we leave?
Final thoughts
Cameras, storage, software, analytics and cloud can be reviewed separately. Replace what is out of support, restricted by procurement, or failing its operational requirement. Keep what still works, and introduce new capability where it changes an outcome. Prove it in real conditions before rolling it out.
One thing rarely makes it into a specification and often decides whether a project succeeds: whether your operators can actually run the system. A VMS that floods the control room with untuned alerts gets worked around within a month.
More on CCTV
Share on: