The term 'embedded vision' has become a much-used buzzword in recent years, giving the impression of a relatively new technology. However, according to Jan-Erik Schmitt, Managing Director of Vision Components, the first standard embedded vision systems were already commercially available in the 1990s: "However, such systems were still called smart cameras or vision sensors back then. The origin of embedded vision systems is therefore the traditional manufacturing industry sector."

However, all experts agree that the main driver for the success of current embedded vision technology is now the consumer sector. In particular, the continuous development of smartphones and the cameras and processors used in them have laid the technical foundation for the rapidly increasing number of applications in many different sectors of industry.

Dr. Christopher Scheubel: "Falling prices and the increasing computing power of processors have made a significant contribution to the success of embedded vision."

© NurembergFair

Dr. Christopher Scheubel, Executive Director of the Framos spin-off Cubemos, sees two opposing spirals that play a significant role in the success of embedded vision: "Because embedded vision is increasingly being used in the consumer market with its high volumes, the prices for the corresponding components and systems are falling. At the same time, the computing power of processors continues to increase steadily."

Another consequence of these trends is that if components with the same performance are sufficient for certain tasks, they are available at a fraction of the previous cost. The increasing success of embedded vision means that more and more of the required components are being used, resulting in further price reductions. For these reasons, the costs of embedded vision systems are amortized very quickly and investments often pay for themselves after just a few months. This aspect also contributes to the great success of the technology. The consumer sector and industry benefit equally.

More applications

Jason Carlson: "Edge computing with platforms such as ARM, X86, GPUs or FPGAs as well as multicore processors provides sufficient computing power for embedded vision."

© NurembergFair

The increasing performance capability of embedded vision solutions combined with falling prices has led to a sharp increase in use cases in many different sectors within and outside industry. This development is not yet complete and, according to the panelists, will affect many more areas in the future that no one is thinking about today.

For Jason Carlson, CEO of embedded computing company Congatec, a key reason for the extreme increase in embedded vision applications is that developments in edge computing are making more and more computing power available: "Images and videos generate large amounts of data and therefore place enormous demands on processing. However, with today's technology and computing platforms such as ARM, X86, GPUs or FPGAs as well as multicore processors, these requirements can be solved better and better and often in real time."

Dr. Christopher Scheubel cites the orientation of driverless transport systems or robot-supported pick-and-place systems as typical industrial applications for embedded vision. For Bengt Abel, Project Leader Technology and Innovation at logistics company Still, embedded vision is also an important enabler for transparent supply and production chains. "There is a large area of application, particularly in intralogistics for transporting goods to production sites, and therefore enormous future potentialfor embedded vision." Abel's company is currently evaluating systems for recognizing people and tracking goods. Embedded vision will also play a decisive role here.

AI and embedded vision

Artificial intelligence methods are currently regarded as a kind of wonder weapon for various areas of application. Expectations are also high in combination with embedded vision - and rightly so, says Congatec CEO Carlson: "Technically, there are many possibilities today to make embedded vision systems based on AI even better and faster. In my opinion, there is only one question that determines whether such solutions will be realized: will they meet the user's expectations at the desired price-performance ratio or not?" According to the experts, the growing range of dedicated, AI-optimized hardware will probably contribute to this technology also being increasingly used in the field of embedded vision.

Dr. Michael Bach: "The development of embedded vision systems with the help of artificial intelligence must lead to increased efficiency for the user."

© NurembergFair

Nevertheless, users must overcome a number of hurdles if they want to use AI methods for embedded vision systems. According to Dr. Christopher Scheubel, a balanced set of data is extremely important in order to achieve good training results. In addition, data labeling - the correct assignment of image content - is manually very time-consuming and therefore expensive, and automatic processes can be prone to errors. The lack of traceability of decisions also makes the certification of AI-based software, which is important for the industry, difficult.

Dr. Michael Bach, Head of Research & Development at CST, explains what users expect: "For us, it is important that AI accelerates the development of embedded vision systems and reduces dependence on the knowledge of individual employees, as this leads to increased efficiency for the user. However, the end result must be a solution that works absolutely reliably and it doesn't matter whether it was programmed with several thousand lines of code or created with the help of an AI system."

PC-based image processing remains

Gion-Pitschen Gross: "Linux-based embedded vision systems offer the advantage over Windows that they can be adapted more individually to the respective requirements."

© NurembergFair

However, the success of embedded vision will not lead to the end of traditional, PC-based image processing, the participants in the discussion are convinced. Embedded vision technology is particularly suitable when a cost-effective solution for a specific application with the highest possible quantities is required. However, if the task requires a certain degree of flexibility, PC-based image processing systems remain the first choice.

In certain cases, a combination of embedded and PC-based vision can also provide optimal solutions. According to the experts, the fact that Linux is usually used as the operating system for embedded vision, but Windows is often used in traditional image processing, is not a problem, as Gion-Pitschen Gross, Product Manager Vision at Allied Vision, explains: "Linux offers the advantage over Windows that the systems can be adapted more individually to the respective requirements. In Linux, you can limit yourself to just the parts of the operating system that you really need. This means that Linux systems can be kept simple and with a small footprint. Linux therefore gives users more freedom to adapt the systems precisely to customer requirements than is possible with Windows."