Innovation might mean seeing a very new path, but that may only require looking at the world with clearer eyes.
Sometimes you think you know the answers to everything. And being a journalist with what some have called a very large ego, I must admit that I fall into this category. Having been around the industry for more than a quarter century (and looking none the worse for wear), I have seen a number of companies develop some very innovative products using some very unusual technologies.
But on a recent trip to Europe, my ego was slightly deflated by what was certainly one of the most innovative ideas I have seen this year. After a whirlwind week-long tour that encompassed England, Switzerland, and Germany, Vision Systems Design’s indefatigable German sales representative Johann Bylek drove me to see Stemmer Imaging in Puchheim, Germany.
After a week on the road with Herr Bylek, I was beginning to feel like Bono promoting the latest U2 album. But that would not deter me from obtaining a story for our magazine. Stemmer Imaging employs some very clever people, and Martin Kersting, technical director, is one of them. But it was not an easy introduction to what they were about to unveil. As we stood over an automated machine designed for credit-card inspection, Kersting asked me what I knew about how to increase the speed of image-processing code used in such machines. And it was here that my ego and lack of experience kicked in.
Of course, I had seen it all before and was only too willing to impart my knowledge (or lack of it) to Herr Kersting. There are only three ways to accomplish this task, I began, sounding like my late father, a lecturer in mechanical engineering. You could run the code on the very fastest processor available or alternatively partition the code on dual- or quad-core processors.
However, although partitioning the code among multiple processors may be effective for companies such as Adobe with a very large installed base, it would probably not be worthwhile in the machine-vision industry, where software volumes are smaller. Alternatively, I expounded, you could allow the system developer to offload some of the most compute-intensive functions as matrix multiplication to an FPGA. Using such an approach would result in a massive increase in throughput, especially as these functions are most often used for tasks such as image filtering.
“And the third option?” Kersting asked with a wry grin. It was obvious and probably one of my best ideas ever: use a dedicated fast-Fourier-transform (FFT) engine to transform the image, perform a point transform instead of a convolution operation, and then use an inverse FFT to obtain the result. Although expensive to implement, this solution was elegant.
At this point I did sound like my father giving a lecture. Martin’s grin became wider. It was obvious he knew something I didn’t. “There’s another way,” he said simply and led me back into the lab where he was working on code. And then Martin posed another question. “What’s the fastest processor in your computer?” he asked. I was stumped at such a simple question. Surely it was the Pentium or AMD look-alike running at around 3 GHz.
That was not the answer. Indeed, the answer was so simple and so elegant that it’s really a wonder I did not think of it first. Today, every computer can be equipped with a very low-cost 16x PCI graphics processor that uses the full 16x bandwidth of the PCI bus. Incorporating multiple floating-point processors in a pipelined fashion, these boards are used to render graphical images with striking reality. Best of all, there is a software standard-DirectX-that allows developers to code graphics functions at a very high-level.
Using pipelined processors, these boards also can be used to great effect to speed image-processing functions. And that (as you will read on page 17 of this issue) is exactly what Stemmer Imaging has accomplished. Now, cameras with no onboard processors can perform operations that may have otherwise been dedicated to onboard DSPs, CPUs, or FPGAs. It is an achievement of which those at Stemmer Imaging can be proud.
Andy Wilson
Editor
[email protected]