• Font Size:
  • A
  • A
  • A

Feature Articles

Trends in Machine Vision Technology for 2007

by Nello Zuech, Contributing Editor - AIA


One can say that over the years the advances in machine vision technology have been incremental. However, when one looks back over the past 35 years, the differences in capabilities are immense. The earliest machine vision systems required minicomputers and they were very limited in capabilities. One of the first companies to recognize the potential of the microprocessor was Object Recognition Systems, my old alma mater. The earliest microprocessors, however, did not have much compute power, so the basic pattern recognition algorithm had to be quite inelegant. The good news was that it was a gray-scale processing algorithm. The bad news was also that it was a gray-scale processing algorithm and it could not distinguish between good gray scale changes and bad gray scale changes. Hence, the number of false rejects was inordinate, unless a fulltime engineer managed its settings virtually continuously.

The other machine vision systems of that era were not that much better. A number of proprietary hardware architectures did emerge to handle more rigorous image processing routines. But, again, these generally worked best with a small set of algorithms, often not necessarily the best algorithms for a given application. The net result of all this early technology was significant advances in the state-of-the-art of the application engineering surrounding the ‘‘staging’‘ for an application – the physical arrangement of lighting, camera and object and especially the lighting design itself as well as interfacing line, lighting and camera. Optimizing the staging was critical to reduce the requirement for compute intensive image processing algorithms.

Fortunately today the machine vision underlying compute technology has made substantive advances over the years. Consequently, more applications than ever can now be successfully addressed. A continuum of configurable machine vision products is now available. Vision sensors with performance superior to the machine vision toolkits that were available 10 – 15 years ago are now in widespread use. Somewhat more intelligent smart cameras in some cases have sufficient intelligence combining FPGAs, DSPs and microprocessors to handle the most compute intensive application requirements. Digital cameras with various connectivity capabilities can readily make a personal computer into a machine vision system. Where a PC may require some augmentation evermore-intelligent frame grabbers are available that can plug into a PC and handle most, if not all, the image processing tasks.

Given the increasing compute capabilities inherent in these products, machine vision hardware based on proprietary designs is declining. More and more application-specific machine vision systems can be handled by one or another configurable machine vision arrangement.

A number of individuals have contributed to this article:

  • Rene Voorwinden – Technical Director – Arvoo
  • Ben Dawson – Director of Strategic Development – DALSA (ipd)
  • Stephane Francois – Executive Vice President – Leutron Vision, Inc.
  • William Munroe – Director of Marketing – Microscan
  • Dr. Lutz Kreutzer – Marketing Manager – MVTec
  • Karl Gunnarsson – Vision Manager - SICK
  • Endre Toth – Director Business Development – Vision Components

What trends are you seeing in configurable vision products (smart cameras, embedded vision processors, PC-based engines, frame grabbers, etc.) that are being used in machine vision?
Arvoo sees as a main trend in imaging hardware the integration of let me say ‘‘camera’‘ and ‘‘processor’‘. A lot of major suppliers do offer 'smart cameras' or 'integrated vision processors', the last with some CCD or CMOS device integrated. We are convinced there is a market for this compact solution, though it is mainly the lower level application market. Besides that, the learning curve for the end-user is very important: DSP or FPGA based smart cameras take their development time for the application programmer who is inexperienced with these specific devices. A solution with a well know operating system as Linux, RT Linux, QNX or Ecos is often preferable, regarding development time and thus time to market.

The main problem for smart cameras that run an OS is that these mostly run on a general-purpose processor such as the Pentium Mobile, Power PC etc. These processors have a high heat dissipation, resulting in a big heat production, inside of the camera!! It is well known that this heat production disturbs a well functioning of the image-forming device, resulting in a loss of precision and a lot of random noise.

For high end applications ARVOO integrated the image acquisition (say: the frame grabber) and the image processing in the axioma™ video processor. The video processor is separated from the imaging device (say: the camera). In this architecture we can offer a high processing power with allowed heat production (about 20W) without corrupting the image forming process. The rack mount solution matches perfectly the high market interest for multi camera applications as 2D and 3D imaging. The Gigabit Ethernet virtual backplane connects multiple units in a system, but allows installing them over a wide area.

[Ben] Machine vision will continue to benefit from the decreasing cost and increasing capabilities of processors, memory, and other components. At ipd, we see three additional trends are increasing markets for machine vision. First, the user interface to the machine vision system is getting significant and overdue attention. The user interface is usually an afterthought, built from the incremental contributions of the algorithm designers. The result might make sense to the designers, but requires long and frustrating study by the user. At ipd, we start with the interface and build it according to accepted human factors principles to make our vision systems easy to use.

The second trend is to make machine vision tools that are task-specific to simplify their use. Instead of a complicated and powerful general-purpose vision system, we make tools specific to a class of problems. This builds in domain-specific knowledge and greatly reduces the expert knowledge required to use a machine vision tool. For example, a traditional machine vision system has dozens of different edge detectors that you might use to dimension a part. This flexibility is good if you know what you are doing, but leaves most users wondering where to start. Instead, we provide dimensioning tools that ‘‘understand’‘ what needs to be done and select the algorithms to do the measurement. We package the vision expert's knowledge into the tool so that the user can focus on their task rather than becoming a vision expert.

The third trend is to increase the intelligence of the vision system so that it can tolerate more environmental variations. For example, instead of requiring the user to ‘‘fixture’‘ or ‘‘stage’‘ parts to a particular location, we use visual search to find the part in the field of view. Or, for example, we use algorithms that better tolerate changes in illumination, so that part lighting is simplified. Combining trends 2 and 3, we have vision systems that are designed for specific tasks, such as inspecting labels. In these cases, the vision system ‘‘understands’‘ the entire task and the interface uses concepts and terms that are familiar and specific to the task. The vision vendor must balance specificity for ease of use against market size and the cost of carrying multiple products.

[Stephane] Technology trends associated with frame grabbers include PCIe and FPGA, multiple GigE Vision™ cameras, image transfer reliability. GigE Vision (and USB 2.0) demand new designs. The market has a desire for solving specific tasks with either do-it-all HW (use in more than 1 project) or task specific HW (easy to integrate for single 1 project). Pricing is always going down for more features.

[Bill] Over the last number of years some of the technologies of machine vision and bar code reading have converged into a new space we refer to as smart cameras or bar code imagers. In this newly converged technology space, there are a number of trends emerging that will likely drive the industry for the next several years. First among these is the growth in the use of digital cameras within the consumer sector. Today you can easily take high quality photos with your cell phone, PDA, or with a super thin camera no bigger than your business card. This shift in consumer acceptance of digital cameras is an indication of a trend I expect to see reflected in the commercial vision sector.

Vision technology will become even more capable, even more user friendly, and also much less expensive. This is similar to what we saw a quarter century ago with the introduction of the compact disc player. When millions of consumers began buying CD players based on laser diode technology, the resulting improvement in the reliability, and dramatic decrease in the price of laser diodes led manufacturers of bar code scanning equipment to move from large, expensive helium-neon (HeNe) laser tubes to the much smaller and less expensive diode lasers. Within just a few years after CD players became available, most all bar code scanners used diodes. The lesson for today? I expect to see commercial vision systems that offer smaller size, superb resolution, and much lower costs. And as plant floor and systems engineers see prices come down for these new vision systems, the number of applications for machine vision should open up.

A second technology trend may possibly prove to be even more important: the dramatic improvement in image processing software tools. While the growth in image processing power and speed has been driven by on-going hardware improvements, better development tools are making it easier and faster for software developers to produce new image processing solutions that can be individually tailored to the needs of a given application.

Much of the physical complexity we have come to associate with setting up a new industrial vision system may be replaced by easy to use software-controllable options. The resulting benefits for plant floor engineers should help open up exciting new applications for smart cameras and bar code imagers.

[Lutz] Some trends that we see associated with the underlying technology include:
Dual Core - essential need for (easy) use of parallel software 64 CPU / 64 Bit XP - full support of 64 Bit for increased speed, more process space, larger images.

Some market related trends that we observe include:

  • Accuracy
  • More interfaces and development of standards
  • Platform independence
  • 3D standard machine vision software for robotics on PC and no-standard-hardware
  • Color displays (e.g., in cellular phone) - Color inspection needed
  • Increasing labor cost - more automation, e.g., in food industry
  • New laws (in Europe) enforce higher quality, e.g., more print inspection in pharmacy
  • Increasing interest in 3D machine vision - intelligent algorithms and high-performance   implementation (high-speed in face of high amount of data)
  • Multi purpose technology machine vision horizontally conquers new markets
  • Regarding this, new users will face machine vision not willing to use proprietary solutions but standard interfaces (GigE Vision, GenICam, etc.); that means components must be exchangeable
  • More standard software
  • More smart cameras
  • Usability: easy-to-use, easy-to-maintain
  • And all-in-all shortening of development time: IDE.

Trends specifically associated with the machine vision software industry?
General requirements:
    1. New Methods for new markets
    2. Speed-up to save money
    3. Ease of use for less skilled people

Trends based on that:
    1. Extensive libraries as the basis for OEM and smart cameras
    2. Continuous improvement of existing code for speed-up:
        + New algorithms
        + Support of 64 Bit
        + Support of MMX / SSE
    3. Based on standard library, development of industry/application
       area specific easy to use tools

A good example is HALCON by MVTec:
A library as HALCON is a perfect product for OEMs, but HALCON Embedded (running on a number of smart cameras and other devices) is an optimal standard software package for machine vision end users. Thus, HALCON Embedded is the most common machine vision software package for special platforms and devices. That means: HALCON is the only machine vision library that is for OEMs and system integrators (available as HALCON
package) and for end users (as HALCON Embedded on special hardware devices).

[Karl] One trend that we are seeing is that customers who have mastered using 2D cameras are now easily able to upgrade to 3D camera technology. With software programming technology that is available, it does not take a huge jump in knowledge to implement 3D vs. 2D. For example, SICK's IVC-2D and IVC-3D cameras are programmed via the same graphical interface (IVC Studio). Monitoring and configuring both cameras is easy. Once configured, both cameras can work in stand-alone mode, without the need for a PC.

However, with 3D, users are then able to gain additional ‘‘height’‘ information, key in many inspection applications. In 2D, for example, a blob tool will capture an area of an object based on intensity of the object. The almost identical tool - when applied in 3D - will get volume of a ‘‘blob.’‘ Volume and area are closely related, the only difference is that volume has a ‘‘height’‘ measurement associated with it. 3D is ideal in low contrast situations or applications where surface features (shape) or volume is of interest, such as in the food or baking industry, where the shape of the product must be uniform (e.g. steaks and chicken filets). Also, applications where it is critical to inspect all aspects of the product, such as pharmaceutical, 3D can further ensure that correct product is placed in the pocket or blister pack. In the future we see 3D vision becoming more commonplace in manufacturing as prices drop and the technology becomes more user friendly.

[Endre] The number of available models has been growing for smart cameras. The different models differ in resolution, frame rates, processing power, all-overall packaging and interfacing and of course in their prices. This makes it possible for users to select the best-fit model for their application to reach the price performance they need. They have real choices to fulfill this goal and take compromises where they see it necessary.

Customers increasingly demand compact systems. For a machine vision task customers do not want to have a huge cabinet, many cables and a few small boxes next to the machine. It is just one of the functions they need in their machine. The turnover for smart cameras shows double-digit growth yearly.





There are currently no comments for this article.

Leave a Comment:

All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.

First Name: *
Last Name: *
Your Email: *
Your Comment:
Please check the box below and respond as instructed.

Search AIA:

Browse by Products:

Browse by Company Type: