Machine vision still needs help

Companies that produce data-acquisition boards and software have made it easy to measure external signals and process them into useful information. Users can display a value, produce a strip-chart graph, or accumulate data for later analysis. Even simple programs let users establish set points and alarms.

Over the years, the suppliers of machine-vision software have taken a similar path. Today's tools make it easy to combine image-processing algorithms without the need to program in C or C++. (But you can still take that route if you wish.) You don't have to know about the inner workings of the algorithms and how to tailor them to specific purposes. Software suppliers' programs let you easily adjust parameters and settings to suit your needs. No programming needed.

But processing images isn't the difficult part of an application. In a data-acquisition system, a user must provide signal conditioning to filter or amplify a raw signal. Vision systems also need "signal conditioning," but it must occur prior to the acquisition of an image. Image signal conditioning involves choosing the proper light source, camera, lens, and optical filters (if any). For machine-vision system developers, there's more to consider than just connecting a few wires and using preset functions.

In the machine-vision world, standard image types don't exist. The products that people must inspect come in all sizes, colors, and shapes. Thus, getting the proper signal conditioning in a machine-vision application usually depends on trial and error. There's no way around it—you have to experiment with lighting, camera positions, and other factors.

Surprisingly few machine-vision vendors offer tools to help people set up the right imaging conditions. Software on Web sites and CD-ROMs helps you evaluate which lens to use and the type of lighting to choose, but overall system setup still remains an art.

Some entrepreneur will start a successful new business by compiling practical information about machine-vision systems and then giving users an expert system to access the data. You could ask such a system, "How do I increase contrast on an SMT board," or, "How do I inspect the inside of a narrow tube?" Answers—or at least suggestions—to those questions exist, but so far, no one has put them in one place in a form that's easy to use.

Perhaps one day the expert system could even anticipate your questions and offer several possible answers depending on how you set up your software. Microsoft took a similar approach with its paper-clip assistant in Word, but most people banished it. "Clippy" just might find a new life in machine-vision software.