Welcome back to our series of exclusive interviews with machine vision CEOs! In this instalment, we are joined by MVTec Software’s CEO Olaf Munkelt.
Olaf tells us how he became a part of the industry, reveals what sets MVTec apart from the competition, and shares how he believes deep learning will continue to evolve in 2024.
1.) How were you first introduced to machine vision?
I have been involved in the machine vision industry for about 30 years. I first came into contact with this technology during my time at university. This led to the founding of MVTec Software GmbH in 1996.
What makes me particularly happy is that we three co-founders are still working together today, albeit in different roles from those we started in. Our industry is extremely dynamic, and this dynamism ensures that I never get bored, even after 30 years.
2.) What is the ethos of MVTec Software, and what sets you apart from the competition?
At MVTec, we develop software for machine vision. Our product portfolio includes the powerful standard MVTec HALCON, the no-code machine vision software MERLIC and the Deep Learning Tool, which enables simple labelling image data for deep learning applications. On top we provide services around our standard products.
MVTec is a family business, and it will stay that way. We have no investors on board, which allows us to plan sustainably and consider the long term. This gives us stability and allows us to continue our chosen course even in uncertain times, without having to achieve short-term financial goals. Our employees also benefit from this business approach.
We also place great emphasis on quality and technology. This is easy to say, but it can be proven. For example, we have our own research department. Here, we work on new technologies and methods. This enables us to provide our customers with the latest machine vision expertise. We ensure the high quality of our products through extensive testing before we release a new version. We are proud of the fact that due to the accuracy of our testing we have very few bugs that appear after the release of our products.
The third point I want to mention here is the hardware independence of our software products. As a software manufacturer we supply one, albeit essential, component for a machine vision application. Therefore, it is very important to us that our software works seamlessly with other components, such as cameras or PLCs. To ensure this, we maintain a large network of partners. Good cooperation is crucial to us on both a technical and personal level.
3.) Which products have brought you the most success, and where are these products most widely used?
It is no secret that we are best known for our HALCON software. Logically, it is also our most successful product. However, with MERLIC and the Deep Learning Tool, we are positioning ourselves more broadly.
With MERLIC, we are targeting machine vision beginners in particular, a constantly growing target group. And with the Deep Learning Tool, we provide customers with low-threshold support for deep learning applications, especially the labeling of data. Deep learning is a relatively new technology, where labeling plays an important role. And with correctly labeled image data, the first successes can be achieved more quickly.
Our goal is to be close to our customers around the world. That is why we do not have a regional focus. We work with a number of strong partners. In some markets, we also have our own presence. In addition to our headquarters in Munich, we have offices in Boston, USA, Kunshan, China, and Lyon, France. Our office in Taiwan, which we opened in April 2024, is brand new.
4.) At MVPro Media, we’re all about improving efficiency in machine vision. How does MVTec Software fulfil this in manufacturing and within your sector?
As a software company, our goal is to help our customers get the most out of any machine vision application. With our software MVTec MERLIC, for example, we provide a no-code software to people with little or no programming knowledge. We are helping them get started with machine vision and automate more and more processes. With MVTec HALCON, we go one step further: this software can be described as a particularly powerful Swiss army knife. It can be used to perform a wide range of particularly demanding machine vision tasks.
Because of its ease of use and easy access to machine vision technologies, MERLIC has been part of the Siemens Industrial Edge platform since late 2023. Within the Siemens ecosystem, users can download the application they need, similar to an app store. Our application, called “Anomaly Detection for Visual Inspection”, helps customers use AI to detect both structural and logical defects in objects, improving the quality of product components. It is easy to learn, test and use, because it does not require programming or in-depth machine vision knowledge. By combining the Siemens Industrial Edge Ecosystem and MVTec’s machine vision expertise, customers can easily harness the power of AI-driven software applications for visual inspection.
One example of HALCON is battery manufacturing. Enormous production capacities are currently being created in this industry. It is also necessary to optimize process steps. With HALCON, which has many powerful functions, we can help. For example, a U.S. company has developed a new laser welding process for bonding cells to the battery case. The welding is done simultaneously by four robots. The exact welding position is calculated and transmitted to the robots by HALCON.
5.) Lastly, do you foresee any especially important trends in machine vision as for the rest of 2024 and beyond?
The most important trend for 2024 will still be the same trend it has been for the past couple of years: deep learning. Deep learning is a method of AI, that enables never-before-seen applications with machine vision in terms of accuracy and effectiveness. Our developments in 2024 will address how to make deep-learning-based machine vision technologies even better.
The new “Out-of-Distribution Detection” feature for deep-learning-based classification that we will integrate into our products can alert users to potentially unreliable classification results. These might be caused by changing environmental conditions, new defect types, wear and tear on the production line, or unexpected circumstances, like wrong or damaged components during inspection. This technology works like an additional test instance for classification by providing users with an out-of-distribution (OOD) score along with the image classification result, indicating whether a similar image has been seen in the training data. A high OOD score means the image is probably not similar to any of the training data, suggesting it may be a new or novel instance that the model has not previously encountered.
The second new feature is “Continual Learning”. This feature solves the problem that networks may completely forget what they already have learned (“catastrophic forgetting”), leading to problems. Continual learning addresses this by allowing users to add new classes or variations to existing ones using just a few images without retraining the entire model.
Dr. Olaf Munkelt is a Co-Founder, Co-Owner, and Managing Director of MVTec Software GmbH. In 1996, Dr. Munkelt and colleagues founded MVTec Software GmbH. Since its founding, MVTec has developed into one of the technologically leading companies in the field of machine vision.
Dr. Munkelt has been a member of the board of the machine vision group of the German Engineering Federation (VDMA) since 2006. From 2009 to 2018 he was chairman of this group. He has been a member of the Board of the VDMA Robotics + Automation Association since 2009.
