Technology by any other name…
Technology that seems like magic in one era can be a machine-vision solution in another.
InProfiles of the Future: An Inquiry into the Limits of the Possible, Arthur C. Clarke formalizes three laws, the third of which states, “Any sufficiently advanced technology is indistinguishable from magic.” More than a quarter-century ago, this advanced idea was embodied in the form of machine-vision systems that relied on racks of processors, memory boards, and proprietary operating systems.
While these systems worked to a degree, they were limited in their ability to fulfill the tasks that they promised. Because of their lack of processing power and high cost, they could only inspect objects at limited speeds in controlled settings, relegating them to high-priced curiosities. Unfortunately, many inFortune500 companies believed the promise of the technology and installed the systems thinking they would replace human operators and reduce the cost of automated inspection.
After these systems did not live up to their promises, they were withdrawn from factory floors, and human operators were reassigned to the inspection tasks. For many years, any project manager who mentioned the words “machine vision” was looked upon with a high degree of skepticism. Machine vision had become a dirty word in the industrial automation business.
One decade later, neural-network technology emerged promising to allow computer-based systems to emulate human brain capabilities. For some, this technology was also indistinguishable from magic. By emulating thousands of neurons in feed-forward networks, it could be trained to classify individual groups of multidimensional variables.
Unfortunately, the lack of processing power of the early IBM PC once again limited the use of the technology. Often, it could take an hour or more to obtain a single result from these software packages. Because of this, the technology was relegated to areas such as credit-approval systems, where a number of weighted variables determine whether a person is a good or bad credit risk.
Today, with the emergence of low-cost personal computers and inexpensive cameras, memory, and networking, the promise of affordable machine-vision systems is finally a reality. And, with its acceptance, many machine-vision hardware and software vendors have incorporated the benefits of neural-network technology into their products. These include smart cameras such as the PCEyebot from Sightech (San Jose, CA, USA; www.sightech.com) and the Zi-640PC CCD ZiCAM from JAI Pulnix (Sunnyvale, CA, USA; www.jaipulnix.com) and software packages such as Common Vision Blox from Stemmer Imaging (Pucheim, Germany; www.stemmer-imaging.com) and MATLAB from The MathWorks (Natick, MA, USA; www.mathworks.com).
These products are not being marketed as a technology, however, but as a solution to specific problems. Rather than highlighting the fact that a feed-forward modified Hopfield network can classify blobs, for example, vendors describe how such systems offer the integrator the capability to build systems “without programming.” Smart OEMs are highlighting the benefits rather than the underlying technology.
These benefits have already proven useful in the food and paper industries, where companies such as Hamey Vision Systems (Sydney, Australia; www.hameyvision.com) and Ventek (Eugene, OR, USA; www.ventek-inc.com) have deployed systems that use the technology. While Harney has applied the technology to define and assess quality parameters for more than 100 products such as cookies (see www.hameyvision.com/biscuit-colour-inspection.html), Ventek’s GS2000 system inspects wood veneer (see www.matrox.com/imaging/news_events/feature/lumbersupport.cfm).
Product vendors who supply OEM peripherals, it seems, have learned the lesson of Arthur Clark’s third law, reformulating it into the more successful, “Any complex technology should not be marketed as such for fear of lack of acceptance.”
Andy Wilson
Editor
[email protected]