The machine vision landscape has already changed considerably in recent years due to powerful further developments in the field of embedded vision - and it will continue to do so. What effects this will have was discussed by high-ranking participants at this year's EMVA business conference in Copenhagen in mid-May. Three board members of the VDMA machine vision division participated.

"Embedded vision will affect us all and become so ubiquitous that there is no way around it". The experts were in full agreement on this key message of the discussion: embedded vision is already so important in the industry today that no machine vision company can afford to exist without this technology and corresponding products in the near future.

What at first glance sounds almost like a threatening statement, has enormous potential, as MVTec Software CEO Dr. Olaf Munkelt explains: "Embedded vision is suitable for a much broader range of applications than traditional PC-based image processing and its main field of application, the industry.” For the existing market and the players active in it the embedded technology thus represents a great chance to think beyond well-known application possibilities and to find applications in which the characteristics of embedded vision systems fully come to validity. This includes, among other things, the peculiarity that the costs of individual components and complete embedded vision systems are considerably lower than those of traditional PC-based architectures, but that development costs are generally significantly higher. For this reason, embedded vision developments usually only pay for themselves from relatively high quantities.

Progress from the Consumer Sector

However, the profitability threshold of embedded vision systems is falling: the current focus for further development of the technology is on the system-on-chip (SoC) approach. To this end, the world's leading hardware and software manufacturers, located in Silicon Valley in particular, are constantly providing new ideas. Companies such as Sony and Intel are already adding more and more new functions to their products at the semiconductor level that can be used for image processing purposes. However, since the underlying technologies are usually developed for mass markets rather than for the machine vision niche, one of the key challenges is to identify these expanded opportunities and leverage existing strengths for effective deployment in appropriate user industries.

A well-known example of this phenomenon has its origins in the entertainment industry: the graphics card manufacturer Nvidia introduced GPU architectures a few years ago. These graphics processors were originally designed to improve the visual representation of computer games. Due to their extremely high computing power, however, these processors were also perfectly suited for the conversion and evaluation of extensive image data such as is available in image processing. Today, the use of GPUs in suitable vision applications is part of the standard repertoire of many developers.

Many experts expect similar effects from system-on-chip solutions, which are currently being developed for use in smartphones and could also be of great value for the image processing industry in the near future. "The development speed of the individual components required for the development of embedded vision systems remains enormously high," observes Dr. Klaus-Henning Noffz, Chairman of the Board of the VDMA machine vision division and CEO of Silicon Software GmbH. "Embedded vision has now reached a state of the art that already enables powerful, economical systems."

Expanding Expertise in Embedded Technologies

In many conversations and technical articles in the recent past, embedded vision has been praised as a kind of miracle weapon that will sooner or later challenge the raison d'être of traditional image processing technology. Participants in the panel discussion did not share this view. One reason for this is the fact that the functionality and performance of conventional vision systems is still increasing strongly at all levels: cameras, image acquisition components, lighting and optical products, software tools for image evaluation, and also the interfaces for data transmission are far from the end of their potential and are being raised to ever higher performance levels by many innovative manufacturers.

The human factor also has an influence on the fact that traditional image processing will be in use for a long time to come: Many programmers with years of experience are available to develop new and further PC-based systems and to create the necessary algorithms. Although the know-how for the creation of embedded vision solutions is growing rapidly, it is not yet available on the market to a comparable extent. One of the challenges for the further development of the embedded vision market therefore also consists in expanding the level of knowledge in companies, e. g. on Linux or Cuda for GPU programming, and transferring the existing application-specific know-how to the new technology. A helpful way of doing this would be to increase the standardization of embedded vision at all levels. In this respect, the relatively young technology is still in its infancy and still requires a great deal of commitment and the comprehensive cooperation of leading companies in order to achieve a status comparable to that of the traditional machine vision industry in terms of standards.

Traditional Machine Vision Remains

The following statement by Dr. Munkelt illustrates the potential changes: "Classic image processing will survive as high-end solutions with corresponding cameras and frame grabbers are still in demand. However, many applications that were considered high-end a few years ago can probably soon be addressed with embedded vision." The limits for using the optimal technology are therefore shifting - but this does not mean the end of traditional image processing architectures.

It is undisputed that embedded vision systems are the perfect solution for certain applications. The experts cited the example of drones, which open up completely new possibilities in applications such as agriculture. PC-based image processing is not suitable for such applications simply because of its size and weight. As a further, economically extremely important field of application, the panel addressed the automotive sector in which the arguments of low price and high volume are the main driving forces behind the development of embedded vision systems. Here, autonomous driving could develop into one of the killer applications that are absolutely necessary for the establishment of a technology. There are therefore numerous opportunities for the use of embedded vision - it is important to track them down and use them!

Embedded Vision at the Embedded World Show: Success Through Collaboration

The embedded world exhibition & conference - 2020 in its 18th year - is the world's leading meeting place for the embedded community. At the embedded world experienced developers come together to share their knowledge and help others to turn ideas and innovations into real products. With over 2,100 participants in the conference and over 32,000 visitors of the show, the embedded world is a well-established platform and market place for embedded technologies in Europe, targeting at developers, project and product managers that are looking for new trends and technologies – people you need to connect with when approaching the topic embedded vision in Europe. VDMA machine vision has teamed up with the embedded world trade fair organizers. In 2020, you can again expect a panel discussion on embedded vision, a dedicated track on embedded vision at the conference, a dedicated award on embedded vision, and many exhibits related to embedded vision at the trade show.