Aoi New Table 1a

Vision Sensors Decide for Themselves

Today's integrated vision solutions can outperform your older discrete system.

You no longer require a Ph.D. in computer science to develop a machine vision system. Now you just need one to understand the product classifications.

The more popular terms include vision sensor, vision system, and smart camera. To these, a few companies have added their own variations such as vision appliance and smart sensor. Some manufacturers distinguish between stand-alone and PC-based vision systems while others maintain that only PC-based solutions or at least systems with separate and powerful processing boxes have the performance to qualify as vision systems.

Although there is considerable diversity in the color-coded definitions presented in Table 1, vision system is at the top of the product type hierarchy. The next most capable product is the smart camera. By definition, all smart cameras are self-contained and do not connect to a PC. They generally are based on a charge-coupled device (CCD) or complementary metal oxide semiconductor (CMOS) imaging device rather than comprising additional circuitry combined with a complete packaged camera.

As Table 1 shows, agreement on a more specific description is lacking. It is clear that the kind of application for which smart cameras are suited is becoming more complex. For example, Machine Vision Products uses a smart camera to perform volumetric solder-paste measurement on printed circuit boards (PCBs). This level of image-processing difficulty lies closer to the high-end fully programmable integrated vision-system interpretation of the smart camera term than it does to the low-end bar-code reader.

Patterned areas are attributes that may or may not be part of a product definition. For example, some Cognex vision sensors have remote head camera that is separate from the CPU. Cognex manufactures both general-purpose and application-specific vision sensors and not all integrate illumination.

Table 1. Comparison of Vision Solution Terminology and Definitions

(for larger image click here)

Vision sensors take third place, with some overlap, behind smart cameras and vision systems. Most manufacturers use the term to mean a machine vision product that can replace other types of sensors. For example, bar-code reading and detecting part presence or absence have been typical uses for vision sensors. This, too, is changing, with some smart camera flexibility being added to allow more kinds of solutions at low cost. A few manufacturers restrict the term to mean only the actual CCD or CMOS element although generally this is called the imaging device or image sensor.

A distinguishing characteristic of vision sensors is that little or no programming is required. Vendors accomplish this by offering limited configuration via a drop-and-drag utility or through a range of application-specific models from which you choose the one most suitable. In contrast, some smart cameras can run the same image-processing algorithms as a full-function vision system. However, most smart cameras can execute only a subset of vision system tools and usually at a lower throughput rate because of limited processing power.

Regardless of the exact definition manufacturers may use, all agree that stand-alone vision-product capabilities are increasing. Customers now have more types of powerful low-cost solutions to choose from.

Distinguishing Among Solutions
Vision Systems and Smart Cameras
RVSI's Jonathan Ludlow, senior director of vision, suggested that functionality similar to that of a PC-based vision system could be provided now in a stand-alone format.  The Hawkeye 1600 is a complete vision system in a smart camera. It has the same machine vision tools as our PC-based solutions. In the past,• he continued,  the necessary camera, lighting, frame grabber, and processing were discrete elements. With the new smart cameras, all of these parts are combined into one piece of hardware, including the lighting and lensing. 

The company's Visionscape• Software provides a point-and-click environment in which you drag and drop appropriate tools into a vision program organized as a sequence of steps. Custom programs can be built to accomplish surface-flaw detection, dimensional gauging, guidance, or identification. In addition, the graphical user interface can be customized using Microsoft Visual Basic or Visual C++. At the highest level, a collection of specific turnkey solutions is available.

The Iris P-Series of programmable smart cameras from Matrox Imaging also combines the space savings of a single-enclosure vision product with the flexibility of a PC-based vision system. This range of cameras uses a field-programmable gate array (FPGA) to preprocess sensor data, releasing the built-in microprocessor to run image processing algorithms and analysis, communications, and control tasks.

 The camera has an ultra-low-power Intel Celeron microprocessor, runs the Windows CE.NET real-time operating system, and is programmed with eMbedded Visual• C++,• noted Fabio Perelli, a product manager at Matrox Imaging.  OEMs and integrators can develop their applications via the Matrox Imaging Library, an imaging processing toolkit, and execute them directly on the smart camera. 

Neural Network-Based ZiCAMs from JAI PULNiX

A third example of a smart camera is the JAI PULNiX ZiCAM, which requires zero programming. The camera learns an application through its neural-network technology known as zero instruction set computing (ZISC), developed by IBM in partnership with Silicon Recognition. Because of its unusual architecture, the ZISC chip provides >250,000 evaluations/s. A conventional Von Neuman processor would need to operate at 2,000 million instructions per second (MIPS) to achieve this level of performance.

According to Mike Kelley, the company's director of Smart Products,  The ZiCAM is finding a place in a variety of niche applications that in the past would have required human judgment. Applications such as sorting fish by species, size, and damage, sorting wood by grain pattern, or inspecting metal by finish, which would have been very difficult for traditional machine vision to handle, are well suited for a ZISC-based smart camera,• he continued.

One weakness of most smart cameras, cited by PC-based vision-system manufacturers, is speed. George Blackwell, the director of product marketing for Cognex PC-Vision Products, explained,  Performance increases with each boost in processor speed, and as a result, PC-based systems often are used for more complex or mathematically intensive applications. However, applying PC-based systems requires more vision expertise and knowledge of low-level programming languages such as C++ or Visual Basic.

 In contrast, vision sensors [Cognex defines everything not PC-based as a type of vision sensor] generally require no programming and provide more user-friendly interfaces. While low cost and ease of deployment remain the key attributes of sensor-based platforms,• he continued,  over the last several years, vision sensors have become increasingly sophisticated while the cost of PC-based systems has come down. 

Compact Vision System from National Instruments

Rather than run custom vision applications on a standard PC, National Instruments (NI) has developed a range of compact vision systems (CVS) that supports image processing and analysis and has ample storage for images and data. Because each CVS platform measures only 4″   5″   2.5″ but accommodates inputs from up to three cameras, the volume required for a high-performance solution is much less than if a conventional PC were used.

According to Kyle Voosen, NI's vision product manager,  Vision systems should be able to perform all of the tasks of a smart camera but with even greater flexibility and speed. Vision systems can be configured either like a smart camera or programmed like a PC. Some NI CVS products offer both methods depending on the application.

 With the greater horsepower offered through vision systems, more algorithm options are available to inspect products,• he elaborated.  For instance, you can take advantage of grayscale filtering to improve an image, pattern matching and geometric matching to locate the position and orientation of components, and classification methods to identify and inspect individual components. While these algorithms are preferred for their accuracy and flexibility, they often are too sophisticated for the simple configuration environments and too complex for the slower processors found in most vision sensors and smart cameras. 

Another advantage of using a separate processor, Mr. Voosen said, is the wide choice of cameras it affords. Although smart cameras often can be provided with a range of sensors, in practice, the number is limited to only a few monochrome and a few color units. In contrast, hundreds of different separately packaged cameras are available.

Steve Geraghty, the director of Coreco's intelligent products division (ipd), added that ipd's NetSight II Integrated Machine Vision System also eliminated the synchronization problems that can be caused by non-real-time operating systems and network-traffic-induced delays. This means that pass/reject decisions are made entirely within the system's 8″   7″   3″ package based on inputs from up to three associated cameras.

Of course, even a PC may not keep up with your high-speed application. Maybe you need a vision processor board to offload image-processing tasks from the host PC.

As an example, the Matrox Imaging Odyssey XCL, based on a combination of the Oasis ASIC and a G4 PowerPC , provides more than 1   1011 operations/s, >4-GB/s memory bandwidth, and 1-GB/s I/O bandwidth via PCI-X. The custom Oasis ASIC integrates a CPU bridge, a links controller, a main memory controller, and a pixel accelerator with 64 parallel processing units.

Vision Sensors and Vision Appliances
Restricted flexibility is a common theme among these types of products. That's not the same thing as low performance. Rather, it's simply recognition that many vision applications involve only one or a few distinct operations.

A Cognex In-Sight Vision Sensor

For example, members of the Cognex In-Sight Series of vision sensors can identify automotive camshafts by color, verify the color of food and beverage packages, and confirm the correct position of color LEDS in electronic assemblies. The In-Sight Series has die-cast aluminum housings, sealed connectors, and an IP67 rating for dust and wash-down protection.

Mark Sippel, product marketing manager for In-Sight Vision Sensors at Cognex, commented,  Because of the compact and scalable nature of vision sensors, they can provide affordable machine vision at each stage of the production process. And, because each sensor is an independent device, the complexity associated with synchronizing multiple cameras off one processor is eliminated.

 It is now cost-justifiable for many manufacturers to add vision in enough places to catch defects at their source. A company that previously had only used vision at the end of the line to inspect the final product,• he continued,  may now be able to justify the addition of dedicated vision sensors upstream to verify and inspect critical steps in the production process. This way, corrective action can be taken on the spot to minimize scrap and the costs associated with rework. 

Vision appliance is the term used by ipd to designate a vision sensor intended to perform a specific task. For example, vision appliances are used to locate a reference mark on a PCB, as described by the company's Mr. Geraghty:

 Because there are no programming costs, a vision appliance is a low-cost solution to many machine vision and inspection tasks. For example, the iGauge Vision Appliance provides optical measurements of features such as point positions, lines, angles, hole diameter, or roundness. It easily can be integrated into a pick-and-place machine,• he continued,  although the tricky part is getting the lighting to provide good contrast for the features of interest. When the task involves alignment and a few critical measurements, an operator or manufacturing engineer can set up a vision appliance in minutes. 

The Hawkeye 1500 Series from RVSI is another example of a vision sensor. The company's Mr. Ludlow explained,  Vision sensors use the same hardware as the smart cameras, but the software is a subset of the machine vision tools. For example, the Hawkeye 1600 Series Smart Camera can run any of the RVSI machine vision tools. The Hawkeye 1500 uses a subset of these tools that only addresses data-matrix reading. The sensor locates, verifies, and reads the data matrix. 

Because all RVSI vision solutions are based on the same Visionscape suite of machine vision tools, only your hardware needs to be replaced should the application change after deployment. So, for example, a vision sensor could be replaced by a smart camera running an enhanced version of the original program.

Omron also has developed a range of vision sensors called ZFV Smart Sensors. The actual sensor is coupled to a separate amplifier that displays the image on a built-in 1.8″ LCD monitor. Suggested uses include measurement of areas, average density, and object dimensions; determination of the presence of a pattern or character string and the edge position of labels or sheets; and alignment of bar codes.

Don t be confused by the lack of uniform terminology. As is readily apparent, a wide range of vision solutions is available and being used. Although the companies listed in Table 1 classify their products differently, they all are machine vision experts and have probably dealt with applications similar to yours. Vision sensors have proliferated, resulting in low cost, which means that one may be the best solution to your problem although in the past it wasn t practical.

If a basic vision product doesn t fit your needs, consider more fully developed vision solutions. For example, Wintriss Engineering manufactures surface inspection systems. In these systems, smart line-scan cameras build continuous images of the surfaces of materials such as copper-clad PCB laminate. The company's Web Ranger• Software coordinates the operation of several cameras in a system with final processing in a PC.

According to Vic Wintriss, the company president,  The system is menu-driven, and no programming is necessary. With the associated Web Analyst• software, menus are generated by quality-control engineers and turned over to production for online use by operators with the Web Inspector• software. A range of high-intensity linear light sources complements the cameras and application software. 

on Cognex vision sensors and
PC-based vision systems

on ipd vision systems
and vision appliances

on JAI PULNiX smart cameras

on Leutron Vision's LVmPC micro-PC
for embedded vision applications

on Matrox Imaging smart cameras,
vision systems, and vision processors

on NI compact vision systems

on RVSI smart cameras

on Machine Vision Products 
3-D measurement system

on Wintriss Engineering
web inspection products


To join the conversation, and become an exclusive member of Electronic Design, create an account today!