Last reviewed
Short answer: AI CCTV is CCTV where software watches the video and recognises what is in the picture, such as smoke, flame, people or vehicles, then sends an alert when something matches the rules you set. Ordinary CCTV records; AI CCTV notices. It can run inside a camera, on a small server on site or in the cloud, and it usually works with the cameras you already have.
Traditional CCTV is a recording system. Cameras capture video, a recorder stores it, and a person watches live screens or goes back through footage after something has happened.
The weak point is attention. Nobody can watch dozens of screens all night. Basic motion detection alerts when pixels change, but it cannot tell a person from a fox, rain or headlights, so it produces more alerts than anyone can check.
AI CCTV adds a layer of understanding. A trained model looks at each frame and labels what it recognises: smoke, flame, a person, a vehicle. Rules then decide which of those matter on which camera, and only those become alerts. The result is CCTV that tells you when something needs a look, with the clip attached, instead of a library of video you search afterwards.
The term covers everything from simple rules on one camera to AI watching a whole site, so it helps to know the main types and terms such as bounding box, edge server and ONVIF.
Real footageMost AI CCTV falls into one of four types, depending on where the analysis runs and what the camera senses. Pick a type to compare them.
The first three analyse the ordinary picture, so they can often use your existing cameras. Thermal is a different sensor. Side by side: video AI vs thermal vs on-camera analytics.
With an edge server, video is analysed on your own network and footage can stay on site. With cloud video AI, streams are sent to remote servers for analysis, so there is nothing extra to install but you rely on your internet connection.
Woodcock AI offers both: an on-site edge server or the Woodcock AI Cloud, chosen per site. See how Woodcock AI connects.
Whatever the type, AI video analytics follows the same basic chain. Here are the six steps on existing cameras.
An existing IP camera or recorder sends a live stream over the network, usually using RTSP, the standard protocol for setting up and controlling real-time audio and video delivery.[3]
Software takes frames from the stream and analyses the picture, either on an edge server on site or in the cloud.
A trained model labels objects and events in the frame, such as smoke, flame, a person or a vehicle, and draws a bounding box around each one.
The detection is checked against the rules for that camera, such as a zone drawn over a gate or a stockpile, so a person in the yard after hours alerts while a parked lorry does not.
If the event matches, an alert is sent by app, email or SMS with the video clip and a short description, so someone can check it in seconds. Alerts to a monitoring centre (ARC) and phone-call alerts are available as optional paid add-ons.
Each detection has a feedback button. Marking alerts as correct or false feeds into retraining, so the AI gets better at what is normal for each camera.
IllustrativeClassification is what separates AI CCTV from motion detection. A motion rule asks whether something changed; a trained model asks what it is. That is why AI can ignore swaying trees and headlights while still flagging a person at a closed gate or smoke rising from a stockpile, and why the rules step matters: a person in the yard is normal by day and worth an alert at night.
AI CCTV is very good at watching lots of cameras without getting tired. It only knows what the camera can see, though, so it is a fast first warning rather than a replacement for alarms or people.
IllustrativeTwo limits are worth repeating. First, video AI sees what is visible: smoke and flame in the picture, not heat building inside a pile, which is where a thermal camera earns its place. Second, it supports people rather than replacing them. For fire, it complements certified fire detection; for security, it makes guards, keyholders and monitoring centres faster, as our comparisons of AI CCTV vs security guards and AI CCTV vs remote monitoring explain.
You rarely need to rip anything out. Most modern IP cameras and recorders share video using open standards, which is what lets AI software connect to them.
The two standards you will hear about are RTSP and ONVIF. RTSP is the protocol used to set up and control live video streams.[3] ONVIF is an industry body whose profiles are fixed sets of features that help products from different makers work together; only products registered as conformant to a profile can be called ONVIF conformant.[1] ONVIF even has a profile, Profile M, for analytics metadata and events such as object classification, covering both cameras and server- or cloud-based services.[2]
In practice, connecting a camera usually means entering its network address and login details so the AI can read the stream. Recorders such as NVRs, and many older DVRs, can pass on streams for several cameras at once. Cameras on a closed manufacturer system, or that do not cover the area at risk, are the usual exceptions, and a free site assessment finds them before anything is installed.
A person can watch one screen at a time. The AI checks every connected view and only asks for attention when something matches.
Woodcock AI works with ONVIF or RTSP cameras and recorders, including NVRs and older DVRs. Installer guide to adding AI to existing CCTV.
A free site assessment checks which of your existing cameras and recorders can connect, which detections suit each view, and whether edge or cloud processing fits your site.
Every site looks different. Steam, dust or road lights can confuse a model that has never seen that view.
Better systems learn each camera. Woodcock AI learns what is normal for each of your cameras, such as natural movement and weather, and every detection in the app has a feedback button. Marking an alert as correct or as a false alarm feeds into retraining, so the system gets better at your site rather than staying generic.
Each alert arrives with the clip, so checking it takes seconds. The answer you give is used to retrain the AI for that camera.
Read more about reducing CCTV false alarms.
Most uses fall into fire, security and safety. A single camera can run more than one detection, for example smoke and fire all day and intrusion after hours.
The sectors that use it most are those with large outdoor areas, valuable stock or a high fire risk: waste and recycling sites, industrial sites, construction sites and vehicle yards. See all sectors or every detection module.
Products vary a lot, and demos are filmed in good conditions. These questions get to what it will do on your cameras.
On the legal side, AI does not change who is responsible. You remain the operator of your CCTV, and the ICO says a data protection impact assessment applies in most cases for video surveillance, with extra care for more intrusive technology such as facial recognition and ANPR.[4] Our guide to AI CCTV and GDPR in the UK walks through the steps.
Real footageAI CCTV is CCTV where software watches the video for you. Instead of only recording footage for someone to review later, it recognises what is in the picture, such as smoke, flame, people or vehicles, and sends an alert when something matches the rules you set.
Mostly. Video analytics is the wider term for software that analyses CCTV automatically, from simple motion rules to AI. AI CCTV usually means a trained model that recognises objects and events, not just movement.
A camera or recorder streams video, usually over RTSP. Software analyses frames on an edge server or in the cloud, classifies what it sees, checks your rules and sends an alert with the clip.[3]
Usually not. Video AI works on the picture your cameras already give, and most ONVIF or RTSP cameras and recorders can connect. Thermal detection is the exception, because it needs thermal cameras. A free site assessment confirms what your cameras can do.
No. Facial recognition is one specific and more intrusive use of AI, which the ICO singles out for extra care.[4] Woodcock AI does not identify people: intrusion detection flags that a person or vehicle has entered a zone you set, and smoke and fire detection looks for smoke and flame.
Number plate recognition (ANPR) is a separate technology. Woodcock AI's vehicle detection recognises vehicles such as cars, vans and lorries in a camera view, where supported, but it is not number plate recognition.
Yes, where the event is visible in the picture, including infrared views. Very dark scenes, glare and heavy fog can still hide smoke or movement.
It depends on the type. With an edge server, video is analysed on your local network and footage stays on site, but a connection is still needed to send alerts to phones. Cloud video AI needs a steady connection with enough upload bandwidth for the streams.
No. It adds an early visual warning to your CCTV and complements certified fire alarms. It can watch every camera at once and alert guards, keyholders or a monitoring centre, but people still decide how to respond.
Yes, when you follow the same data protection rules as for any CCTV, including a clear purpose, signs, sensible retention and, in most cases, a DPIA.[4] Our AI CCTV and GDPR guide explains the steps.
Standards and guidance referred to in this guide. This guide is general information, not legal advice.
A free site assessment reviews your existing cameras and recorder, the risks on your site and whether edge or cloud processing suits you. Support is available 24/7. Alerts to a monitoring centre (ARC) and phone-call alerts are available as optional paid add-ons.
Illustrative 3D render
