Definition
Computer vision is the field in which software interprets images on its own: recognising what is in the frame, where it is and what it is doing. In industrial use this means turning a camera stream into a record of events that can be acted upon, instead of footage somebody has to watch.
How it differs from ordinary CCTV
Classic CCTV records images and waits for a person to review them, usually after an incident, while establishing what happened. Computer vision evaluates the image as it arrives and reports only deviations from defined rules.
The difference is practical rather than technical. An operator watching a wall of monitors loses attention after tens of minutes, while a system evaluating every frame has no better and worse days.
What a vision system consists of
The first layer is the image source, that is the cameras already installed on site. The second holds the detection and tracking models. The third, usually missing from discussions about artificial intelligence, is the rule engine describing the specific plant: zones, schedules, exceptions.
Next comes the response layer - notification, signalling, an entry in the ticketing system - and an event archive with evidence. Without the last two, detection remains a curiosity with no effect on the statistics.
Where it works and where it does not
It works where an event is visible and repeatable: missing protective equipment, entry into a zone, a pedestrian approaching a vehicle, smoke, a spill, an object present where it should not be.
It will not replace sensors where the phenomenon is invisible: gas concentrations, temperature inside a machine, bearing vibration. A well-designed system combines the two instead of pretending a camera is enough.
Frequently asked questions
Is computer vision the same as artificial intelligence?
No. Computer vision is an application area, and current solutions in it are built on machine learning. Classic methods without learning also exist and are used for simpler measurement tasks.
Are special cameras required?
For most safety applications ordinary IP cameras are sufficient. Specialised cameras, thermal ones for example, are needed where the phenomenon is not visible in the visible spectrum.
Does the video have to be sent to the cloud?
It does not. Analysis can run locally on a device at the site, which removes network latency and the question of footage leaving the premises.