Vision System Online Help

Machine vision applications involve widely varying degrees of difficulty. Some are straightforward and easily handled by one of the so-called smart cameras. Smart cameras have sufficient built-in processing capabilities to perform basic image analysis tasks. Vision sensors are similar but more limited in the kinds of applications they address. Both types may provide a complete solution by integrating lighting, a camera, image processing, and camera-to-host I/O.

More difficult applications require some outside help unless you have experience designing and implementing machine vision solutions. Many system integrators with the necessary experience are happy to discuss possible approaches to your problem. You need a clear understanding of what must be accomplished, but you don’t have to know how to do it. This is the point at which contacting machine vision experts makes sense.

What about the large range of applications between the relatively easy and the very difficult? What are your chances of handling the jobs yourself? How much online information is available? What can you learn on your own about machine vision even if a system integrator ultimately is called in?

Online Resources

On many websites, the most common form of user help is a selection matrix. In most cases, these are product-oriented rather than application-oriented.

A camera selection matrix might include the sensor size, vertical and horizontal pixel resolution, frame rate, and perhaps interface type. Unless you have a reasonable level of experience in camera selection, it’s not easy to make trade-offs among a large number of parameters when presented this way.

Technical notes, case studies, and white papers also can provide useful information. Not surprisingly, white papers and case studies generally deal with a company’s own technology and products. Given the expertise a company may have in something like smart cameras, you can learn a great deal about capabilities and deployment by reading a few case studies.

Broader, less product-specific information is presented in a technical note. A popular topic on several sites is comparison of CMOS- and CCD-based cameras. In fact, this topic has been a popular subject for many years. Initially, CCDs were by far the technology of choice, having better dynamic range and linearity. Today, CMOS sensors have improved to the point that for many applications they offer better performance at a lower cost.

Many companies publish their own online newsletters. Although this document generally is used as a marketing tool, newsletters often contain helpful application stories. Even if the problem addressed isn’t exactly the same as yours, the information still can be useful. Prosilica, PPT Vision, Matrox Imaging, and Cognex are a few of the companies that regularly publish this kind of free material.

Training programs are offered by many suppliers, sometimes conducted at your company and sometimes online. A few introductory courses are free, but generally you have to pay for training. Specific questions or problems can be dealt with by technical support engineers. Websites always give an e-mail address and sometimes a phone number for technical support. In addition, many sites list frequently asked questions and answers.

Interactive design tools present a vendor’s products from the user’s point of view. Lens selection programs are a good example. In many of these, you are required to enter approximate dimensions for the field of view, working distance, and the smallest dimension that must be imaged. The number of camera pixels and the size of the sensor also affect lens choice. However, it’s very easy to enter values for which no lens is suitable, and selection programs that return “no lens found” are of little help.

Instead, a few selection guides return the closest match. This is a great help to the novice machine vision developer who needs to become familiar with the interactions among the many parameters.

For example, if you wanted to capture an image of a 1 m x 1 m object but with only a 100-mm working distance, a very large angle of view is implied. The closest match in this case might be a lens that required a 5-m working distance. Immediately, you can see that this is the specification causing a problem.

It’s a matter of perspective—no pun intended. A product-specific selection matrix focuses on the available products, and you simply need to choose among them. In contrast, an interactive selection program that shows you the closest match is focused on your application: Tell us what you need to do, and we’ll see how close we can come to satisfying the requirements.

In addition, because choices often are not independent, you need to understand the limitations implied by the rest of the vision system. For example, a major benefit of GigE Vision™ cameras is that a separate frame grabber isn’t required. Instead, camera data is transferred via gigabit Ethernet directly to the host PC.

While this is an economical approach, it requires host processor bandwidth that may not be available. And, the performance of a real-time vision system can be affected by both the Windows operating system and a standard Ethernet network interface card (NIC). Software drivers that bypass the host operating system improve performance by directly accessing memory. Also, some special hardware incorporates frame grabber functionality and reduces host overhead, especially for multiple camera systems.

Once an image has been transferred to the host, it needs to be analyzed. Vision software toolboxes from several manufacturers allow you to address a wide range of applications without being a programming expert. Often, you can download a demo version for evaluation.

A particularly interesting approach is offered by SILICONSOFTWARE, in which users string together VisualApplets^® to define a series of image operations. The VisualApplets actually are compiled modules based on lower-level operators from the company’s image processing library. You use the company’s synthesis program together with Xilinx place-and-route software to generate an FPGA layout for your custom machine vision system.

Noteworthy Online Help

Technical Notes
The Viewbits Division of Computer Modules lists several cameras and states that CMOS sensors now are at least as good as CCD sensors. The site references an article on CMOS vs. CCD written by Rockwell Scientific.

In it, the author explains that the SNR of CCD sensors has an inherent 3-dB/octave decrease as the readout rate is increased. In a CCD, charge is transferred from well to well during the readout process so each well must have a much higher bandwidth than is required to capture the image.

According to the paper, the output from high-definition sensors with a high frame rate often is followed by a 30-MHz to 33-MHz filter to reduce noise. The very wide bandwidth of the CCD output amplifier is largely responsible for excessive thermal noise.

In contrast, especially with submicron technology, CMOS sensors can separately amplify each pixel and have a much lower bandwidth. “The noise bandwidth [is reduced] from tens of megahertz for CCDs to tens of kilohertz for CMOS [sensors].”¹ This means that the SNR associated with CMOS sensors can be much higher than that associated with CCD sensors used at high output data rates.

CCD sensors are a mature technology so their production will not stop abruptly. Nevertheless, for many applications and especially where a high pixel rate is needed, CMOS sensors are performing better. Most importantly, CCD sensors require special silicon processing that’s not compatible with usual CMOS fabrication. Nevertheless, large CCD arrays used in lower-speed applications such as astronomy have excellent performance with SNR of approximately 120 dB.

On the Melles Griot website, you can find the Machine Vision Lens Fundamentals technical note. It covers important aspects of lens operation and terminology. This is a good reference in general and especially useful for anyone just starting to work with machine vision. After reading this paper, at a minimum, you will know which questions must be answered before you can choose a lens. Because a large number of lens properties are discussed together with their effects, some of the trade-offs within your application should become more apparent.

Edmund Optics^® provides a great deal of lens-related information on the company’s site. Often, you can learn the basics from a well-illustrated paper written in plain English. For example, at the end of the white paper Will a Telecentric Lens Optimize My Machine Vision System? you can select more detailed telecentric lens information. The white paper acts like the top level of an illustrated menu.

Dalsa is another good source of technical information. A series of technology primers looks particularly interesting, with titles such as 5 Critical Factors for Choosing Digital Image Capture and FPGAs Accelerate Machine Vision. In addition to white papers on cameras and sensors, there are approximately 50 application notes on frame grabber applications.

Prosilica makes many types of both Firewire and GigE cameras. Detailed information on each model is available on the website; however, there doesn’t appear to be a concise selection matrix. Nevertheless, on the plus side, clicking on support on the home page immediately gives you access to approximately 25 documents that deal with all aspects of camera selection and software.

Although this entry might better be described as an online book rather than a technical note, Fundamentals of Image Processing is a technically detailed description of most aspects of image processing. The text covers topics from the Fourier transform to edge detection from a mathematical point of view.

ImagingProcessingPlace.com provides a list of 52 tutorials related to image processing. They deal with a wide range of topics, many of which directly relate to machine vision.

A general word of warning is in order when accessing any online technical information. Although some material truly doesn’t age quickly, many cutting-edge aspects of 1999 technology aren’t interesting in 2008. Datasheets have to be updated or rewritten to correspond to new product introductions, but white papers and technical notes on some websites have been allowed to get very out of date.

Selection Matrix
The Vision Components website lists six camera families with 25 models in total. Basic specifications are given including sensor size, resolution, frames/s, shutter speed, processing capability, number of digital I/Os, and the type of digital interface. If you have an idea of what you need, this might be sufficient information to help narrow your choice of models. A small picture of each type of camera helps distinguish models, those with built-in lighting or OEM versions without enclosures as well as more conventional types.

Basler has a camera selector that appears easy to use. When you move your mouse over the various selections, icons representing the available cameras stay dark, but the unavailable ones gray-out into the background. It’s easy to tell as you browse line rate, resolution, and color or monochrome whether there is a camera available that can address your combination of parameters. Basler places a lot of emphasis on interfacing because the company makes both GigE Vision and Camera Link cameras.

The JAI site lists several cameras and has a good selection matrix. Rather than changing intensity or color as Basler does, the JAI matrix is redrawn each time you make a selection. So, having started with all the types of cameras presented, the monochrome ones disappear when you select color. Only the color ones are then presented and so on as you work your way through scan rate, interface type, video standard, and light spectrum choices. The result is much as with the Basler matrix, just accomplished in a different way.

Leutron Vision has several cameras available, distinguished by their interfaces. There are five types, and you can list all the cameras in a matrix of features or create a smaller matrix with only those cameras having a certain interface.

Navitar’s Optical Wizard is a lens selection tool that could be very helpful. From the specified object size, working distance, and camera sensor details, the Wizard determines the lenses that will satisfy or come close to meeting your requirements. The results include several parameters and may be split into two sections if there are conflicts among the input data.

For example, a solution may satisfy the specified working distance but not give the needed object coverage. Or, the object may be covered but at a greater working distance. Both solutions are given together with pictures of the types of lens suggested.

The PIXCI^® selection guide on the EPIX website is impressive simply because it’s so large. It lists many camera manufacturers and hundreds of models together with several of their basic specifications. For each, you can determine the level of EPIX XCAP software support and the most suitable frame grabber. Because it is comprehensive, the listing may be of use to you, but it will be more informative if you already have a good idea of the camera you would like to use.

Matrox Imaging also features frame grabbers. Under the main products heading, there is a hardware product selector. You first indicate whether you are using a digital or an analog camera and whether its interface is Camera Link, GigE Vision, IEEE 1394a/b (IDC), LVDS/RS-422, or Serial Digital Interface (SDI). A further clarification such as Camera Link with or without power may be required, leading to the final step where suitable frame grabbers, vision systems, and stand-alone systems are listed.

Analysis Downloads
VisionPro from Cognex is configured via drag-and-drop, scripting, or programming. Drag and Drop in VisionPro QuickBuildâ„¢ lets you configure acquisition, optimize vision tools, make pass/fail decisions, communicate with the factory floor, and create operator interfaces.

The Programming Toolkit supports Visual Basic, C#, and managed C++. Many code snippets and dozens of sample applications cover a wide range of uses.

The VisionPro Tool Library features the PatMax application for geometric pattern matching and inspection and IDMax to read data matrix codes and verify printed characters. Several utilities add the capabilities to identify and sort images by color and find and measure parts of images. You can download the trial software and use it free for 30 days. At least 400 MB of free disk space is required.

With Data Translation’s Vision Foundry, you create image processing, inspection, and analysis scripts by selecting a suitable sequence of tool icons. In addition, using Microsoft Visual C++, you can develop new tools and through a custom script tool modify Vision Foundry-generated scripts. Vision Foundry supports FireWire and USB cameras as well as the company’s PCI frame grabber boards.

You can download and run a trial copy of Vision Foundry free for 14 days. This is a 92-MB file. Online help includes a Vision Foundry tutorial, the user manual, and several technical papers related to working with image data.

Tutorials
Banner Engineering has extended its traditional photosensor technology into the smart sensor area. Rather than one or a few photosensors being involved, the image from a camera is analyzed to determine pass-fail. An online tutorial consists of three sections: products, theory, and Q&A. Banner is one of few sites that actually uses the term Online Learning as a topic heading on the home page.

Automated Vision Systems offers a free online Fundamentals of Machine Vision course. It requires no prior machine vision experience, takes about 1½ hours, and is accompanied by an audio commentary.

Matrox Imaging has several types of training courses. The only one listed as free is Part 1 of working with the company’s FPGA Developer’s Toolkit. Instruction is available as a five-hour classroom course or a free webcast. If you already are familiar with the Matrox Imagine Library but need to offload processing tasks to an FPGA, this course may be for you. Part II is not free.

National Instruments (NI) regularly organizes free user meetings in major cities throughout the United States and abroad. Machine vision may be part of a larger discussion or can be the focus, such as the Machine Vision and Scientific Imaging Hands-On Seminar recently held in Texas. Image processing software is a big part of the NI story, but the company also has developed a range of smart cameras and backs up hardware and software products with years of PC-based machine vision experience.

You can download NI’s Vision Resources Kit for free. In addition to information about the company’s smart cameras, the kit contains a white paper about choosing a machine vision system, an image processing tutorial from NI, and a very detailed and well-illustrated imaging primer from Edmund Optics.

Application Evaluation
If you would like to contact a machine vision company to ask for an engineering evaluation of a proposed project, the form you can download at the PPT Vision site is a good place to start. When completed, this document identifies you and your company; describes the vision project and its purpose; and includes sufficient budget, environmental, and operational constraints that a meaningful proposal can be developed.

In addition to more usual forms of online information, Edmund Optics supports a discussion forum. Users around the world post questions that are answered by Edmund applications specialists.

Cognex also offers a free application evaluation service.

Summary

Nineteen companies associated with machine vision have been highlighted. Many of them offer a great deal online, and all of the websites are worth a visit. We don’t claim that this Web survey is comprehensive, but it does indicate the type of information that is readily available. It’s up to you to make use of the material, perhaps starting with a subscription to one or more newsletters.

Reference

1. Kozlowski, L. J., “CMOS vs. CCD: Changing Technology to Suit HDTV Broadcast,” Rockwell Scientific, HDTV.videotechnology.com.

FOR MORE INFORMATION
on the fundamentals of image processing
click here

on digital image processing books and educational resources
click here

December 2008