Some CMOS cameras deliver 1280 x 1024 SXGA images at 26 fps, then switch to deliver 320 x 240 images at 340 fps. This allows for maximized frame rates with minimized data to acquire and transfer across PCI busses.

When human eyes are too slow or costly to screen, machine vision keeps better track of flying production lines and tight tolerances. And since their first trials in the semiconductor industry twenty years ago, vision systems have evolved considerably. Also called automated optical inspection, constant monitoring of objects can now extract complex information from images and generate sophisticated decisions.

The first machine vision function is inspection such as measuring machined part dimensions for unacceptable tolerances, often right on production lines. The second machine vision use is tracking; barcodes and matrices affixed to products or boxes are scanned for quick identification. The third vision application is guidance. Once a vision system has processed images, inspection results are used by a waiting controller. (For example, a system might automatically make measurements and then guide a robot arm into place.) A machine vision system often performs a combination of these functions: Parts might be moved into position for the detection of blemishes, scanning, sorting, and positioning (to help a picker grab them) before they’re moved along to makeway for more parts.

CCD cameras

Vision begins at the sensing device. The most prevalent design, CCD cameras are charge-coupled devices developed in the 1960s. The wafer, often with metal oxide semiconductor capacitors, is a solid-state electronic component segmented into an array of individual light-sensitive cells called photodetectors. (Because photodetectors are elements of the picture whole, they’re also called pixels.) CCD pixels sense incoming light by the photoelectric effect, the action of certain materials to release electrons when hit with photons of light. As long as light impinges on the pixels, electrons (fenced in by nonconductive boundaries) accumulate in them. A good industrial CCD has a pixel array of 640 x 480 or so, with each pixel less that 10 microns square.

The row-by-row processing of data in a CCD is where the sensor gets its name. Arrays of polysilicon electrodes — one for each pixel — are packed close to allow travel of electrons to other electrodes. A sequence of clock signals shifts charges across this chip surface towards an amplifying register that measures charge. Like a bucket brigade, a row of information is transferred to the readout register while following rows are shifted closer. After being fed out, electric charges are released and the register emptied for the next row in line, until all information is read.

A drawback to CCDs is that to prevent signal degradation, clock pulse amplitude and shape must be controlled; this requires a clock chip with multiple power rails. CCDs also suffer from blooming — where charge leaks from one photodetector into others. Finally, the step-by-step process of a CCD is not conducive to high speeds. However, a solution for this exists: In higher-end frame interline transfer CCDs, readout registers as large as the light receptor area read absorbed information in one pass.

CMOS cameras

Commoditization is partly responsible for the spread of machine vision use, and CMOS cameras could keep this trend going in the near future. CMOS, or complementary metal oxide semiconductors, is actually a generic term referring to the process of making CMOS devices including not only cameras but also computer RAM, processors, and other semiconductors. Because these cameras are made with the same equipment as other common components, economy of scale equates to cost savings.

Used in digital imaging, active pixel cameras integrate separate charge amplifiers at each pixel. Performance increases while noise levels fall, but (because each pixel needs at least three transistors) more costly silicon area is used. Active support circuitry is located near light receptors to cancel noise right at the pixel. CMOS pixel fill factors are lower than with CCDs, because extra circuitry bordering each pixel takes up space. And though the problem has been addressed in some designs, image quality of CMOS can suffer from fixed pattern noise caused by unequal amplifiers at pixels. What are the benefits of CMOS, then? Processes can be integrated right into CMOS chips, eliminating circuitry for analog-todigital conversion, clocks, and white balancing. This translates into reduced power consumption — sometimes to just one-third that required for comparable CCDs. Another benefit over CCDs: If only a small section of an image is of interest, it can be accessed directly.

PCs and frame grabbers

To record color information, a CCD camera must have red, green, and blue filters over individual pixels to make each sensitive to only one color; then pixels return charge values for its color only. Besides this RGB method, another HSI sensing method exists.

After images are collected, a computer must process and make conclusions about them. Though other systems are growing, the everincreasing flexibility and performance of personal computers have ensured their place in complex designs and replaced proprietary systems. Even standard-issue PCs can juggle information from multiple cameras. On these PC-based systems, most often an interface board or graphics card (also called a frame grabber) converts camera images into digital data. The computer, in turn, uses that data to make decisions about the manufacturing environment. But besides converting and transferring data, frame grabbers compensate for poor lighting, varied camera data styles, and simpler optics while interfacing with conveyors, lighting systems, and rejection mechanisms through I/Os. While frame grabbers were once very application- specific, newer designs are more expandable and generic for changing requirements.

There are some drawbacks to PC-based systems running non-realtime operating systems. Their inability to prioritize communication — especially when multiple cameras are involved — is problematic on high-speed and safety applications. Explains Jayson Wilkinson, Motion Control Product Manager at National Instruments Corp. in Austin: “Windows doesn’t prioritize tasks as well, so things like the mouse could throw the whole system off. But solutions exist: Realtime environments ensure that assigned tasks take place in a deterministic amount of time. In some development environments, engineers can even develop code in familiar Windows and then download that code to a real-time target. And now, some small stand-alone units can connect multiple Firewire cameras to ensure that vision tasks are given highest priority.”

George Blackwell of Cognex Corp., Milwaukee, explains, “The development environment allows users to build (set up, and program or configure) vision applications to meet specific needs. PC-based systems have a programmable environment and offer the most capable vision tools. They also provide the fastest performance because they rely on the latest CPU architectures and as a result, they are generally used for more complex or mathematically intensive applications.” However, Blackwell also points out that applying PC-based systems requires more knowledge of low-level programming languages such as C++ or VisualBasic.

Continue on Page 2