Adding AI to the cameras you already have

The camera estate on most sites represents years of spending: cameras, cabling, recorders, poles, maintenance contracts. When someone suggests AI analytics, the fear is that it all has to be replaced. Usually it doesn't.

How analytics reach an existing camera

Almost every IP camera and network video recorder sold in the last decade can stream video over the network, usually using a standard protocol called RTSP. Analytics software connects to that stream the same way your recorder does, watches it, and reports what it sees. The camera doesn't know or care that something new is watching.

There are two common setups:

• An edge box on site: a small computer on the local network that reads the camera streams, runs detection locally, and sends only events (and the clips that go with them) out to the cloud. Good for sites with many cameras or limited upload bandwidth.
• Cloud processing: streams go directly to the analytics service. Simpler for a small number of cameras on a good connection.

Either way, your existing recorder keeps recording exactly as it did before.

What to check first

Can you get a stream? Find the camera or recorder's RTSP address, username and password, and check a stream opens in a player like VLC from a machine on the same network. If it does, analytics can use it.

Is the view any good? Analytics can't fix a poor view. Check each camera for:

• a clean lens, and no spider webs
• a sensible angle: people should be more than a few dozen pixels tall where you care about them
• no pointing straight into the sun or floodlights
• usable night performance, with working infrared or enough lighting

What's the network like? An edge box needs to reach the cameras locally and the internet outbound. Cloud processing needs enough upload bandwidth for every stream you send. Sub-streams at lower resolution are usually plenty for detection.

Which cameras matter? You rarely need analytics on every camera. Start with perimeters, entrances, yards and the places incidents actually happen.

Older and analogue cameras

Analogue cameras connected to a DVR can often still be used if the DVR offers network streams, and many from the last ten years do. If not, an encoder can convert analogue feeds to network streams, and that's still much cheaper than replacing cameras.

Keeping what already works

A good analytics setup sits alongside your existing system:

• the recorder keeps recording
• the existing monitoring centre can still receive alarms, now filtered to the ones that matter
• the cameras, cables and poles stay exactly where they are

The upgrade is in what happens to the video, not in the hardware that captures it. For most sites that means a much smaller project than people expect: an afternoon to connect, and a week of tuning.