Custom Systems Keep Changing With the Technology

Ask a layman what a custom machine vision system is and they're as likely to point to a complex inspection task with vision systems based on PCs, add-on imaging processing engines and specialized cameras - assuming they have some inkling about what machine vision is!

The layman would be both right and wrong. The fact is that very few machine vision systems sold today are not customized in some fashion. 'Every system is custom in a sense,' explained Vic Wintriss, president of Wintriss Engineering Corp., a company specializing in web inspection and high-performance cameras (San Diego, CA). 'Obviously, what you'd like to do is build a machine vision system that will do every job for every customer. And just as obviously, that's impossible…but the days of 'I'm going to build a special system for you and nobody else' does not make sense anymore. Every application is custom in some way or another, but companies cannot afford to build a unique solution for every application. So what we do is, every time a customer comes in and needs something special, we throw it in the mix and make it available for everybody. And by now we pretty much have any tool for web inspection that a person would ever need. '

Host vs. Smart Camera
Rob Schoenberger, president of Agris-Schoen Vision Systems Inc. (Alexandria, VA), admits that defining a 'custom machine vision system' is a difficult exercise. 'Basically, you have smart camera and PC-host based approaches. While you will customize a smart camera application, I would not really call that a custom system, while PC host systems are often custom systems. However, today a low-end camera with firewire or USB output and PC can cost less than a smart camera in some cases. If you have a PC and can run a piece of off-the-shelf software and not make any substantial changes or customization, it's not really a custom system either,' Schoenberger said.

Schoenberger further defines a custom system by what the system has to accomplish. While some would define 'high-speed' applications like web inspection as an application requiring a custom system, many times these applications can be accomplished using off the shelf components, again letting he air out of standard 'custom system' definitions. Rather, Schoenberger suggests that the components themselves have a lot to do with what is a custom system. 'The need for advanced communications is one way to define a custom system. If you have to talk to an Oracle database and then choose a type of inspection based on the result, that often leads to custom systems. Smart cameras can do it, but often not as well. If you can do it with smart cameras, then it's probably still at the low-end of the applications.'

Other common system requirements that prompt a custom solution are the need for cameras that run above the normal 30 Hz and require double buffering or pipelining, or applications that require more than the standard 8-bit depth. Applications requiring fine color differentiation in the presence of texture changes are also candidates for color systems because there is a dearth of color image processing libraries available today that address fine color identification, Schoenberger said. Also, Schoenberger added, 'By definition anything that requires a custom algorithm is by definition a custom system, or a [less than adequate] integrator. It can't be that you just chose a custom algorithm, but that a custom algorithm is your only option.' One example is an application that requires multiple ROIs with low contrast within the ROI, but extremely high contrast between the ROIs [image dynamic range], requiring more than one grayscale stretch or mapping of the histogram.

Customized smart cams
Part of the difficulty in defining a custom system and the applications that require them resides in the quickly advancing technology of machine vision. What used to require a $250,000 investment 15 years ago can be done better today with a low-end PC or even a smart camera. In fact, more and more of the PC is finding its way into the smart camera. For instance, Wintriss' new Opsis 500 FPS uses a single board computer with Ethernet output and FPGA for high-speed blob analysis at rates up to 500 fps for the full array, or 500,000 lines per second if windowed to 1 line. The CMOS sensor itself can be 'windowed' seven ways from Tuesday - and all these functions would have required a high-end VME based system just a few years ago.

Market Intelligence News & Insights: 

How Will Apple’s $500 Billion Investment for US Manufacturing, Education, and AI Play Out?

On February 24^th, Apple announced a “commitment… to spend and invest more than $500 billion in the U.S. over the next four years”, with plans for new manufacturing plants, enhanced support for manufacturing partners via their Advanced Manufacturing Fund, and a myriad of other initiatives. As part of this plan, Apple projects 20,000 new hires, with roles focused on R&D, silicon engineering, software development, and AI/ML engineering.

Read More

'Smart cameras have augmented the sensor market, and exposed manufacturers to the capabilities of machine vision,' commented Mario Mancini, president of Applied Machine Vision Inc. (Pittsburgh, PA). 'But every machine vision system out there requires some kind of customization. Machine vision is an exciting, scalable technology. By that I mean, that as the application increases in complexity, the more customized the system.'

According to Mancini, there are only a few smart cameras that are designed for a single purpose, such as some bar code and 2D symbology readers. 'There are off-the-shelf products out there do that do part presence, label verification, etc., where you can just set it up, train it and it's a go. But as you get more detailed in your application, such as making specific measurements from a particular point, those are things that you'll have to configure out of the toolset you're using. The more complex the requirements, the more customization you'll need.'

To Mancini, choosing the right toolset from the products available on the market is an important form of customization that integrators provide. 'There are a lot of vendors with analysis toolsets, and some are better than others. One will have strong color toolsets, another strong geometric pattern finding toolsets. That decision process itself is a form of customization, and that's what integrators bring to the table…sure, there are cases when a person could buy something off the shelf and get it to work initially, but what about future changes in operation? How will the system fit best into an overall operational plan? Those are custom considerations too,' Mancini observed.

Software and the moving line
Toolsets and software are among the most visible ways to customize a vision system. Although most would agree that lighting and sensor selection can make or break the image processing analysis, algorithms will always be a critical part to a successful machine vision system. 'After you pick the right toolset, you still have to figure out which tools apply to a particular application,' explained John Salls, president of Applied Vision Systems (Minneapolis, MN).

'I've never seen an application that couldn't be solved with machine vision, but sometimes it can require someone with a PhD in lighting and computer science to get the job done. That aside, you can solve 90 percent of applications with standard [image processing] toolsets. If you can't, then it may not be worth pursuing. Of course, I've seen a client spend $200,000 on programming and $80,000 on hardware for a very custom, complex system. Based on what the system could do, it was worth the investment,' Salls said.

If history has taught the machine vision industry anything, it is that change is the only constant. In another five or 10 years, that $200,000 system that filled a 5-foot rack will probably fit on a Palm Pilot. 'The dividing line between smart cameras and high-end PC systems is constantly moving. The line today is very different from where it was 5 years ago, and it will be very different from where it will be in five years,' commented Agris-Schoen's Rob Schoenberger.

All agree that image analysis is moving closer to the camera all the time, blurring the faint line between an off-the-shelf solution and custom system. 'With a dumb camera, you need incredible bandwidth to get the images back to the processor. Think about it, 90 percent of the image is thrown away. You're only interested in the ROI, so why send the entire picture back along a cable? In the not too distant future, they're going to have Pentiums on the same chip as a CMOS sensor and that'll be that,' said Vic Wintriss.