Machine vision: A view of trends
January 1st 2009

Machine vision has become a mainstream technology, with the majority of applications moving out of the realm of the specialist and into the domain of the production or process engineer.Mike Bailey, systems engineer for Vision and Automation, National Instruments UK & Ireland, comments

Machine vision is taking place all around us; products we use will have been inspected with it, positioned by it, counted by it and identified by it. During the past 10 years, I have seen the acceptance of this technology grow to the point where I am no longer talking about something strange and alien to many engineers.

Machine vision is evolving and growing to become easier for the majority of applications, and configuration-based interfaces are certainly helping this trend.

Users of vision systems no longer need to be hardcore C programmers to get the performance from a top-end PC to work in a real world application. Certainly PC speeds and capabilities have improved and helped this along, but by no means have production capabilities and rates remained stagnant either. Configuration-based interfaces allow users to try a variety of tools very quickly and find their way to a solution. This leads to rapid development cycles and results can be obtained much more quickly. The understanding among users that vision is not a simple sensor has also helped. By adding a vision system to a production line, end users can learn a lot more about their production capabilities and quality.Watching a production line for a day will only ever let you know about that day's production, and will only show a tiny slice of the variations of products and situations produced. The ability to log images and postprocess them is invaluable as it allows users to adjust their inspection routines to suit the subtleties of their production process.

No matter how easy a vision system is to set-up, it has to be affordable and show a business benefit. The good news is that the overall cost of vision systems has dropped, with this being attributed to a combination of new camera technologies and connectivity. Standardisation of cabling is an important factor.We have seen a move from analogue and parallel digital (parallel digital being expensive since virtually every cable was custom made), to the standards of Camera Link, IEEE1394 (Firewire) and GigE. IEEE1394 and GigE have the additional advantage of not requiring a frame-grabber card since the cameras can be directly connected to the PC motherboard. Some systems have progressed beyond this to incorporate camera and processor into one box. The smart camera is a growing trend that will continue. The ability to check production at the right point, cheaply and easily, makes great business sense. Defects and wrong parts can be handled or rejected at the point of failure before any further value is added or production time wasted.

With many systems getting cheaper and easier to configure, it would be easy to reject the need for vision specialists but it is evident that high-end vision systems are also growing in popularity. As production techniques and products get more complex, coupled with the ever growing quality demands of today's society, ways for inspecting products must match this. High end inspection typically requires high end computing power. Camera buses and PC buses allow us to take in huge amounts of data, Camera Link will allow up to 680MB/s, virtually a CD of images a second, through a PCI Express framegrabber.

This, if you consider a megapixel image, gives you a processing time of less than 2ms. There is very little processing that can be done in this period of time.

Fortunately, multicore and multi-processor technologies can help.

Vision engineers are now forced to make use of techniques and programming environments that allow them to more easily target multicore systems. This allows them to use many processors in a system to perform different parts of the application, or even split the image into smaller parts for processing across the cores in a system.

How this is done, is where the expertise is required, but the programming of this must be, and as is being, made simpler for users.

Multi-threading, thread management and data synchronisation will cause even hardened C programmers to crack their knuckles before getting stuck in.

Algorithms and compilers that automatically manage these complex tasks for users are the way forward. Multi-core is something that will not go away and will work its way into smart cameras as chip makers are releasing small, low power multi-core systems that will benefit these systems.

Vision is moving main stream and the change can be attributed to three main contributors: Non specialist engineers can now more easily design, configure and deploy customised vision applications; overall system implementation cost has significantly reduced; and the supercomputing ability on the desktop machine provides the power to deal with the vast amounts of required processing.

More articles from National Instruments UK Limited: