northdan.
Vezi pagina în română

IT Glossary

What is Computer Vision?

The branch of AI that teaches computers to see: recognizing objects, text, faces and defects in images and video.

To a computer, a photograph is just a grid of numbers — computer vision is the branch of artificial intelligence that turns the grid into meaning: here is an invoice and these are the amounts on it, this product has a scratch, that is a car entering the parking lot and this is its plate number. The technology has climbed down from the laboratory straight into ordinary operations: automatic reading of invoices and receipts for accounting, visual quality control on the production line, shelf inventory from photos, identity-document checks at customer onboarding, people counting in retail spaces. The entry threshold has collapsed — cloud services sell “sight” through APIs at fractions of a cent per image, and modern models can be specialized on your particular case with hundreds, not millions, of examples. Rules worth engraving before a project: define precisely which decision the “seeing” supports, test on your real images (bad lighting, crumpled paper), and leave uncertain cases to human validation — good automation is the kind that knows when to ask.

Let’s talk about your project

Message us on WhatsApp or send an email — you talk directly to a developer.

office@northdan.com · +40 752 070 247

Why it matters for your business

Eyes that never tire

Visual checks — quality, documents, stock — run at the same accuracy on piece ten and piece ten thousand, nonstop.

Paper becomes data on its own

Scanned invoices, receipts, delivery notes and IDs turn into structured records — the dullest copy-paste in the company disappears.

Insight from cameras already installed

Existing video feeds can count visitors, watch zones and flag events — the infrastructure is paid for; it was only missing a brain.

Frequently asked questions

How accurate is automatic document recognition?

On standardized, decently scanned documents, accuracy routinely exceeds 95 percent; the remainder is handled by the right workflow — the system flags uncertain fields and a person validates only the exceptions. Manual entry, for comparison, has error rates of its own, plus the cost of the hours.

Can we use computer vision with our existing security cameras?

Usually yes — the analysis is added in software, on top of current video streams, locally or in the cloud. Mind privacy law, though: anonymous people counting is one thing, facial recognition quite another — the latter demands a serious legal basis and an impact assessment.

What data do we need to train a model on our specific case?

Less than a few years ago: for many tasks, a few hundred labeled images (good part / defective part) are enough for a pilot, because modern models arrive pre-trained and merely specialize. The small, measurable pilot beats the mammoth project.