A peek at tomorrow's machine vision

Nov. 11, 2004
Senior Technical Editor Dan Hebert puts forth the idea that future vision systems should be organic, invisible parts of the control loop and operation of the machine, and not just an add-on at a later time.
MACHINE VISION SYSTEMS are one of the fastest growing areas of the control system marketplace. ARC Advisory Group (www.arcweb.com) predicts a compounded annual growth rate (CAGR) of 8.9% over the next five years. ARC says the market hit $860 million in 2001 and should exceed $1.3 billion in 2006.

The approximate size of the machine vision market can be predicted with some certainty, but it is much harder to predict how technology will affect and improve vision systems. For this task, we asked the vision system vendors for their take on what vision systems might look like in five years. We also asked how these advances could translate into new applications.

One consensus view is that complementary metal oxide semiconductors (CMOS) imaging chips will replace traditional CCD sensors, greatly improving resolution and reducing costs. "As CMOS imager performance improves, high resolution will essentially be free," says Nello Zuech, president of Vision Systems Intl. (www.imagelabs.com/vsi). According to Steve Gieseking, director of R&D for DVT (www.dvtsensors.com), CMOS imaging chips will enable machine vision companies, especially "smart camera" manufacturers, to substantially lower prices.

We have been hearing about artificial intelligence (AI) for decades, but the technology must be ready for prime time since Steven Spielberg made a movie about it. Mark Sippel, vision product marketing manager for Omron Electronics (www.ormon.com/oei), feels AI will have the greatest influence on the future of machine vision. "Today, a user must still have a path to the application solution in mind to begin the process of setting up the product," he says. "But what if the user only knew what he wanted to accomplish, and had no idea how to set up the vision system?"

According to Sippel, a user could tell an AI-enabled vision system to check for features, make dimensional checks, or search for surface flaws. The vision system would then look at the image or images of several parts and make the necessary determinations of what algorithms and tools to use. After an automatic "learning phase," the vision system would configure itself to perform the desired tasks.

The vision system would automatically set built-in lighting to achieve the best image and contrast. It would also inform the user of any manual changes needed for the system to perform at an optimal level. "The continued advancement of hardware is a given in today's world," adds Sippel. "Software advances to incorporate AI technology would make machine vision systems truly easy to use."

Lighting is one of the more complex areas of vision systems, a place where automatic learning could greatly increase ease of use. "Light sensing and processing systems will be able to adjust automatically for environmental changes," observes Endre Toth, director of business development for Vision Components (www.vision-comp.com). "The processor will be able to set the lighting and the objective according to the current environment and the object to be analyzed."

Toth also foresees use of different frequencies of the spectrum. "Vision systems will use the wavelength that shows the investigated features most naturally and distinctively," he adds. "More and more machine vision systems will use multispectral sensing and processing, scanning a number of frequencies and combinations to provide the best results and most detail in the analyses."

A more prosaic, but equally important, technology advance is occurring now in the area of communications. "Ethernet is the emerging standard. We are seeing Profibus over Ethernet (Profinet), Rockwell's EtherNet/IP protocol, and Modbus TCP/IP supported in most Ethernet I/O devices," says Gieseking of DVT.

As technology advances in any area, new application areas are created because of lower costs, higher performance, and better packaging. Jason Mulliner, vision product manager for National Instruments (www.ni.com), predicts that neural networks will allow for more classification-type applications, many of which will probably find their way into machine vision. He thinks biometrics will greatly benefit from this technology as facial recognition, iris detection, and other biological identifications become commonplace.

Mulliner thinks that as processors become faster and algorithms more efficient, using machine vision for traffic guidance applications could greatly decrease energy consumption. "Imagine all cars on a road equidistant to each other and responding to every movement in a deterministic fashion," says Mulliner. "Stoplights would become less of a nuisance and highway gridlock would become non-existent."

Machine vision systems traditionally are fixed systems performing inspection tasks detecting after-the-fact phenomena. According to Toth, future machine vision systems will be "organic" parts of the machine, almost invisible to users. Toth says these organic sensors will be part of the control loop and operation of the machine, and not just an add-on at a later time.
About the Author

Dan Hebert | PE

Dan Hebert is a contributing editor for Control and Control Design.