btools.hu

Machine vision & AI

Teaching machines to see.

Classic vision systems on Cognex and Keyence, model-based AI vision, our own image processing software and image annotation: from the first sample image to inspection on the line.

OK 0.99 SOLDER 0.94 OCR LOT 2611 CAM-01 · 12 MP · 2.1 ms ● LIVE MODEL v3.2 · ONNX · GPU OK 1284 NOK 7
EtherNet/IP
OPC UA
99.4 % yield

Two approaches

Classic and AI-based machine vision.

Rule-based inspection where precision and speed matter, deep learning where defects vary too much for hand-written rules.

Rule-based

Classic vision systems

System design and programming on Cognex and Keyence platforms: choosing camera, optics and lighting, calibration, PLC integration.

  • Gauging, positioning, robot guidance
  • Presence and absence checks
  • OCR/OCV, 1D and 2D code reading
  • Surface defects, colour and pattern checks

Deep learning

Model-based AI vision

In-house image processing software and neural networks, from labelling the images to running them on the production line.

  • Classification, object detection, segmentation
  • Anomaly detection from few defect samples
  • Dataset building and image annotation
  • Edge inference (GPU, ONNX), PLC/MES integration

Applications

What we inspect.

Surface defects

Scratches, dents, stains, cracks and coating faults on metal, plastic, glass and textiles.

Assembly verification

Every part present, in the right place, the right way round, before the next station.

Measurement

Calibrated 2D dimensions, gaps, angles and positions at line speed.

Codes & text

1D/2D codes, DPM codes, OCR and print quality verification for traceability.

Robot guidance

Finding parts and passing positions to the robot for picking and placing.

Sorting & classification

Grading natural or variable products where simple rules are not enough.

AI vision pipeline

From first image to monitored model.

  1. Capturecamera, optics, light
  2. Labelannotation, QA
  3. Trainmodel, augmentation
  4. Validatemetrics, thresholds
  5. Deployedge, PLC link
  6. Monitordrift, retraining

A good model starts with good labels.

We also do image annotation and dataset management: precise, reviewed labels, consistent classes and versioned datasets.

Try it: the image shows a scratch, a dent and a stain. Pick a class, then draw a box around that defect. We score your box against a reference label, the way annotation quality is checked in real projects.

Drag with mouse or finger

Feasibility check

Not sure it can be done? Send us images.

Send a few images of good and bad parts. We tell you whether classic vision or AI suits the task, what camera and lighting it needs, and what accuracy you can expect.

Start a feasibility check

FAQ

Common questions.

How many images do we need for an AI model?

It depends on the task. Anomaly detection can start from a few dozen good images; a defect classifier usually needs a few hundred labelled examples per class. A feasibility check tells this to you immediately.

Classic vision or AI: which do we need?

Measurement, code reading and presence checks are usually best with classic tools: fast, deterministic and easy to validate. AI pays off when defects vary in shape, size and appearance. Many systems combine both.

Can the system talk to our PLC?

Yes. Results go to the PLC over EtherNet/IP, Modbus or OPC UA, and images and results can be stored for traceability.

Do you take on projects where the task is limited to image labeling?

Yes. We label images for your own AI projects with reviewed quality and the class definitions you need.

Have a project in mind?

Tell us about it, and let's find the right solution together.

Get in touch