Mr. Paintner, could you please briefly outline the status quo of today's image processing solutions?

Michael Paintner: For many years, image processing was the domain of specialists: from camera technology and lighting to data transfer to specialized hardware, the technology was complex. Then the large volumes of data also had to be evaluated - a challenging task for software developers, whose algorithms first and foremost had to be very fast. Ultimately, the systems - for example in quality assurance - often only need to output a simple signal as to whether the current part is OK (pass/fail information).

However, due to the complex technology stacks and, above all, the high engineering costs, such solutions were often very expensive. In the meantime, the hardware has become significantly cheaper - but the engineering is often still very complex. This is where we focus our efforts in order to reduce complexity for the customer.

Why does the topic of user-friendliness play a major role in the field of machine vision in particular?

Smartphones, which have become an integral part of our daily lives, are a good example of how successful a high level of user-friendliness can be. Configuration and operation have become so simple that they can be carried out by practically any user. And this is precisely the approach that we are also pursuing in machine vision: Users don't need to be experts in machine vision or have any programming skills!

Based on our history as a sensor manufacturer, we try to ensure that our image processing solution is as easy to configure and operate as a classic sensor.

Vision sensors are already considered to be much easier to use than cameras. What solutions does ifm offer in this regard?

We have been pursuing the smart vision sensor approach for many years. It doesn't matter whether it's a laser line scanner such as the new 'Profiler', the 'Dualis' series for code, blob and contour detection or a 3D camera from the 'O3D' series, which determines a distance value for each pixel - in all cases, intuitive and fast configuration and commissioning of an application are the most important goals. In this way, image processing solutions can be scaled in ever new areas. And in future, this will also make it possible to make further progress in applications with autonomous mobile machines - we will be offering new solutions in this area over the course of the year.

How do you manage to avoid requiring the user to be a specialist in image processing when setting up and evaluating the captured images?

The key lies in the software: our 'Vision Assistant' is easy and intuitive to use, even without detailed image processing expertise. With the new 'Dualis', for example, we have already mapped all common inline quality control applications as well as automation tasks such as robot guidance via coordinates in the software. The corresponding image processing algorithms are already programmed and can be used directly. All the user has to do is select and configure the right application for their case - without any programming effort. From setting the lighting and focusing to outputting the results, the application can be put into operation and checked with just a few clicks.

And where does the journey continue?

The interview partner: Michael Paintner, ifm "Users don't need to be experts in machine vision or have any programming knowledge!"

© ifm

I have already mentioned the platform for autonomous mobile machines that we will be launching on the market later this year.

Another focus will be the use of artificial intelligence and deep learning. Our 'ifm mate' worker assistance system is a good example of what is possible in this area. This worker assistance system is based on recognizing the human hand in a live image. A task that is extremely demanding and differs significantly from conventional pattern or contour recognition as used in classic image processing solutions. A hand can be clenched into a fist, its palm can be facing upwards or downwards, the fingers can be outstretched, angled or spread - and in all cases, the system must be able to reliably identify the hand in real time.

We use powerful deep learning technology to solve this task. With this artificial intelligence method, a neural network learns automatically through experience. The system was trained with images of hands until the identification worked reliably. Here, too, we rely on systems that the user can put into operation in just a few minutes without any prior knowledge. And we will continue with this approach in the future.