Definition
Computer vision is the field of artificial intelligence that enables software to interpret images and video: recognising objects, reading text, detecting defects, tracking movement and understanding scenes. It turns pixels into structured information a program can act on. Computer vision powers everything from the camera that reads your car's number plate at a car park to the scanner that digitises your receipts and the quality camera on a production line.
In business, computer vision shows up in manufacturing (defect detection), retail (shelf monitoring, self-checkout), logistics (package scanning, damage detection), agriculture (crop monitoring with drones), security (access control) and back office (document digitisation). It has matured from bespoke, expensive projects to off-the-shelf capabilities, and modern multimodal language models can now handle many computer vision tasks with a plain-language instruction instead of a custom model.
Computer vision is one of the oldest successes of deep learning and remains essential where speed, cost or precision matter, for example on cameras that must decide in milliseconds without a cloud connection. Multimodal models have absorbed the general-purpose cases. The misconception is that computer vision needs perfect lighting and huge labelled datasets. Pre-trained models and synthetic data have made small, practical deployments realistic.
In practice
A food producer installs a computer vision system that inspects every package for seal defects at line speed, rejecting faulty units automatically and reporting defect rates to the shift manager in real time.
Why it matters
Any process where a person looks at something before deciding, from inspection to document handling, is a candidate for computer vision. It is often the fastest route to measurable savings on the operations floor.
Frequently asked questions
- What is computer vision used for in business?
- Quality inspection in factories, reading documents and invoices, monitoring shelves and stock, verifying identity, detecting damage in logistics and insurance, monitoring safety compliance on sites, and analysing medical or agricultural imagery. Anywhere a person watches or reviews images, computer vision can assist or automate.
- Do I need a custom model for computer vision?
- Less often than before. Multimodal language models handle many tasks such as reading documents or describing images with simple instructions. Custom or specialised models are still the right choice for high-speed inspection, precise measurement or devices that must work offline.