Machine vision

A camera only takes an image. A vision system combines an industrial camera, the right lighting and software that analyses the image and makes a decision itself, for example approving or rejecting a product. The system immediately passes the result on to your machine, PLC or operator.

We test it first. In-house we have the equipment to determine, with samples of your products, which camera and which lighting give the best result.

For complex or innovative projects we carry out a preliminary study, so you know in advance which components are needed and how feasible the project is. Want to see it in black and white first? Request a Proof of Concept.

That depends on the type of camera, the distance from the camera to the product and the size of the product. For every project we analyse which type of camera is needed and which underlying software we need.

A line scan camera captures one line of pixels at a time while the product moves past underneath. Line by line, this builds a sharp image of the entire surface. We use them for fast-moving or very long products, such as paper, film, textiles, sheets or products on a conveyor, and for cylindrical products that rotate in front of the camera. When you need to detect small defects across a large width, a line scan camera is often the best choice too. When we combine the camera with a laser line, laser triangulation also lets us measure the 3D profile of the product, such as height, flatness or thickness. For stationary or smaller products, a regular area scan camera is usually enough.

We decide that together with you. The operator can get an immediate notification, the product can be rejected automatically or the line can stop. In addition, all results can be stored, so you can check afterwards what happened.

We can configure the system to inspect different products. We then create a recipe for each product in the recipe library. At a product changeover, you select the right recipe or teach the system a new product by presenting one correct reference product.

Traditional vision software works with fixed rules and is ideal for clear features such as codes, dimensions and positions. For natural products or defects that look different each time, an AI model learns by itself what is good and what is not, based on example images. We make sure this happens in a controlled way in our vision software.

Deep learning is a form of AI in which a neural network learns by itself to recognise what a good or a rejected product is, based on a set of example images. It is particularly strong for defects that are hard to capture in fixed rules, such as scratches, stains, colour deviations or natural products that look different every time. We start by collecting and labelling images of your products, train the model and test it thoroughly before it goes into production. After that, we can keep retraining the model with new images, so detection becomes ever more reliable.

For custom vision systems, we work with HALCON, the vision software by MVTec. You need one licence per industrial PC, and we can connect several cameras to it. It is a lifetime licence: you pay once, without annual subscription fees. That keeps costs predictable, even when you add extra cameras to the same PC later on.

The camera locates where and how a product lies and passes that position on to the robot or cobot. This way the robot or cobot can always pick up the material in the right position and then place it correctly in a machine or on a pallet.