Can You Add AI to the Cameras You Already Have?

Usually some of them. Rarely all of them. The honest answer depends on one thing that has nothing to do with how old the cameras are.

It depends on whether the picture contains enough usable detail for software to make a decision from.

The short answer

AI analytics can run in three places: on the camera, on the recorder, or on a separate server fed by the cameras. If your cameras cannot run it themselves, which most older cameras cannot, analytics can often be applied at the recorder or server instead, using the video the existing cameras already produce.

That means the realistic question is not “are my cameras AI capable”. It is “is my footage good enough to analyse, and is my network capable of carrying it”.

What actually determines whether a camera can be used

Four things, in rough order of importance.

Pixels on target. Analytics needs enough detail across the subject to classify it. A camera covering a wide yard at 1080p may have plenty of pixels across a person at ten metres and almost none at forty. The same camera is therefore suitable at one end of its view and useless at the other.

Image quality at night. This is where most retrofits fail. A camera that produces a usable daylight image and a smeared, noisy, heavily compressed night image will generate unreliable detections precisely when you need them. Motion blur from a slow shutter in low light is a particularly common cause.

Frame rate and compression. Aggressive compression saves storage and destroys the fine detail analytics depends on. A camera streaming at a low bit rate to save disk space may need its settings changed before it can be analysed, which in turn increases storage requirements.

Angle and position. Cameras mounted high and pointed steeply down were often positioned to observe activity rather than identify people. Analytics generally performs better on a flatter view of a subject. Sometimes a retrofit is mostly a repositioning job.

Age matters less than people expect. A well-specified camera from several years ago, mounted correctly, in reasonable light, is frequently more usable than a cheap recent one in a difficult position.

Where the analytics actually runs

On the camera

Modern cameras with built-in analytics classify objects themselves and send events rather than raw video.

Advantages: low network load, no central hardware, usually the most accurate because the camera analyses an uncompressed image.

Disadvantages: requires replacing cameras. Analytics are tied to each camera’s capability.

On the recorder

Many current NVRs can apply analytics to connected cameras, including older ones.

Advantages: no camera replacement. One upgrade covers multiple cameras. Usually the cheapest route to a meaningful improvement.

Disadvantages: limited by the quality of the video the camera sends and by how many channels the recorder can analyse at once. Check the channel limit, which is frequently lower than the total camera count.

On a separate server

A dedicated appliance or server receives streams and analyses them.

Advantages: most flexible, platform independent, scales furthest, and generally the route for licence plate recognition and more advanced analytics across larger sites.

Disadvantages: highest cost and the heaviest network load, since full video has to reach it.

A realistic upgrade path

Most Melbourne commercial sites do not need to choose between keeping everything and replacing everything. The useful middle path looks like this.

Step one: audit what you have. Which cameras produce a usable image at 2am, at the distance that matters, at the angle they are mounted. This is a site visit, not a spreadsheet exercise.

Step two: fix the few that matter most. There are usually two or three positions that carry most of the risk, typically a vehicle entry, a main pedestrian door and a high-value storage area. Replace those with cameras suited to the job.

Step three: apply analytics centrally to the rest. Recorder or server-based analytics on the existing cameras that are good enough, accepting that some will be detection only rather than identification.

Step four: add verification where it pays. Live monitoring on the positions where an intervention would actually prevent loss.

This is how the Mordialloc wheel and rim specialist site was brought together: Hikvision CCTV, Akuvox access control, Elite Alarms and FS Vision operating as one system rather than four separate ones. The published case study also records the arrangement extending to Adelaide. The detail is in the wheel and rim case study.

What AI analytics genuinely delivers, and what it does not

It genuinely delivers:

• Far fewer false triggers, because the system distinguishes a person from a possum, a vehicle from a moving shadow, and rain from motion.
• Searchable footage. Finding a white van between 11pm and 1am becomes a query rather than an evening.
• Licence plate recognition at vehicle gates, which turns a camera into a record of who arrived.
• Alerts that are worth responding to, which is the precondition for anyone actually responding.

It does not deliver:

• Perfect accuracy. Every system has false positives and missed detections, and anyone claiming otherwise is selling.
• Recognition of individuals by identity, unless you have deliberately implemented that and dealt with the substantial obligations that come with it.
• A replacement for positioning and lighting. Analytics amplifies a good camera position and cannot rescue a bad one.
• Compliance. Switching analytics on creates documentation obligations before you do it, which is covered in AI cameras and the 2026 privacy rules.

The network question nobody asks

Server-based analytics means full video streams crossing your network continuously.

On a small site that is unremarkable. On a site with twenty cameras at 4K, it is a meaningful load, and it will share infrastructure with everything else the business does. Three things to check before committing:

• Whether the switches can carry the traffic and power the cameras.
• Whether the cameras sit on their own network segment, separate from business systems.
• Whether the recorder is reachable from the internet, and whether it should be.

That last point matters more than most buyers realise, and it is covered separately in the article on the network behind your cameras.

Common mistakes

Assuming age decides it. Position, light and lens matter more than the year on the label.

Upgrading the recorder and expecting every camera to improve. Central analytics is limited by the worst stream it receives.

Ignoring the channel limit. A recorder advertising analytics may support them on only a subset of its channels simultaneously.

Turning analytics on and leaving the defaults. Detection zones, sensitivity and schedules need tuning to the site, or the system produces either noise or silence.

Switching it on before documenting it. The compliance step comes first.

What to check before you call anyone

• Pull up last month’s night footage from your three most important cameras. Can you identify a person? A plate?
• Find the model numbers of your recorder and a few cameras.
• Check whether your recorder’s analytics channel limit covers all your cameras or only some.
• Note which positions you would replace first if you could only replace three.

That short list makes the first conversation far more productive than a general enquiry.

Frequently asked questions

Will AI work with cameras from a different brand to my recorder? Often, through standard protocols, but feature support varies and analytics support is the first thing to be limited in mixed-brand setups. It needs checking per combination rather than assuming.

Can I keep my existing cabling? Usually yes. Cabling is frequently the most reusable part of an older system and often the most expensive to replace.

How many cameras do I need to upgrade to see a difference? Fewer than most people expect. Upgrading the two or three positions that carry the real risk usually produces more benefit than upgrading ten low-risk positions.

Does adding analytics increase my storage requirement? It can, because analytics often needs a higher bit rate than a storage-optimised configuration uses. Factor this in rather than discovering it later.

Can I start with one camera? Yes. Starting with one high-risk position and expanding is a legitimate approach, and it is how several sites are best handled.

Where to start

The point of an upgrade is not newer equipment. It is that when something happens, you can tell who it was, and ideally that somebody noticed while it was still happening.

FS Vision is designed to work with the cameras, alarms and intercoms already on site rather than requiring a rip and replace, and it can begin on a single high-risk camera and expand later. A free on-site assessment establishes which of your existing cameras are worth keeping and which two or three positions are actually carrying your risk.

Book a free on-site security assessment or call 1300 232 152.