CitationDownload as .RIS
Emerald Group Publishing Limited
Copyright © 2005, Emerald Group Publishing Limited
A stroll in the park
A stroll in the park
My commercial career started about 25 years ago with the foundation of the UK's first machine vision company, Electronic Automation. At that time the vision systems we tried hard to sell were low resolution and pretty slow. They were also regarded as expensive and very much an “unknown quantity”. Performance was erratic as the crude image processing algorithms struggled hard to work reliably under varying lighting conditions.
Since then things have changed considerably and machine vision is now so robust, low cost and easy to use that the question raised by most assembly system designers is not “shall we use machine vision?” but instead “what can we inspect and how many cameras will we need?”. You still need to be careful when designing systems, but lessons have been learned and are well documented.
The machine vision market is on an upward curve with sales growth of around 20-25 per cent per annum (various sources). According to Cognex it has sold over 225,000 systems since it was founded in 1981 and this amounts to over 1.7 billion in US$ revenue. These are impressive statistics, but to my mind they serve more to underline the future potential than to show how well we have done to date.
As our viewpoint by Professor Roy Davies makes clear the bottleneck is no longer with processing speed but is instead with the way we use the processing power that is available to us, and how easy it is teach a vision system what we want it to look for. I fully agree with Professor Davies that the human vision system may well provide us with clues regarding a better way of doing machine vision.
If you read a specification for a machine vision system you will be bombarded by seriously impressive numbers regarding instructions per second, mega pixel resolution and specialised signal processing instructions. These have given us the sledgehammer potential to do automated visual inspection at very high speed. And as time progresses I have no doubt that these areas will continue to be improved and further progress will be made.
The big question though, is whether or not we are actually going about machine vision in the correct way. It is a bit like travelling from “St James Park” to “Victoria” by heading east on the London Underground Circle Line. Over the years you make the trains go faster and by careful sequencing you minimise delays. You might even introduce sophisticated acceleration platforms so that passengers could alight without the train needing to slow down or stop at all.
But then one day someone looks at the problem in a new way and realises that “Victoria” is only one stop away if you travel west instead of east, and what's more you can get out of the hole in the ground, take a stroll in the park, feed the pigeons and still arrive at “Victoria” before the east bound train.