Camera concepts come clear
Knowledge of specs and operations helps you choose a digital camera.
Chris Brais, Coreco Imaging, Intelligent Products Div., Billerica, MA -- Test & Measurement World, 11/1/2004
Digital cameras have pervaded many consumer and commercial applications. Because of the ease with which they can be connected to computers along with the resistance of digital signals to noise in industrial environments, these cameras have also become worthy contenders in inspection systems. Choosing a digital camera, though, requires an analysis of how they work and how to apply them.
 |
| Figure 1. Area-array sensors come in many sizes to fit various imaging applications. The on-chip electronics capture an image in one exposure. Courtesy of Kodak. |
Generally, CMOS sensors don't offer the low noise performance of CCD sensors, but they provide benefits that include low cost, high-speed operation, high pixel density, and low power consumption. In addition, CMOS sensors resist "blooming," or a leakage of charge that occurs in overexposed CCD sensors. The leakage saturates surrounding pixels and often ruins portions of an image. (Blooming doesn't damage the CCD, though.)
Cameras that employ CMOS sensors also offer another advantage: They can access individual detectors and obtain only a portion of an image or a region of interest. Reducing the image data gathered at its source increases the speed at which a camera can acquire images and lets image-processing tasks take less time.
Pixels line up
 |
| Figure 2. A line-scan sensor relies on the motion of an object to provide sequential lines of intensities that result in an image. Courtesy of Kodak. |
Software in a host computer combines the line-by-line information to construct a 2-D image. Line-scan cameras inspect moving PCBs, rotating cylindrical objects, and long products that might otherwise require several area-scan cameras.
The resolution available from line-scan cameras usually surpasses that offered by area-scan cameras. A 4-kpixel line-scan camera is fairly common, but a 4-kpixel-x-4-kpixel area-scan camera is not, and its cost would far exceed that of a line-scan camera with a similar pixel count.
As a rule of thumb, you should use an area-scan camera when you have a stationary object or an object that moves only slightly during inspection. Use a line-scan camera for objects that move in one dimension or for applications that require the inspection of a long product that could require multiple area-scan cameras.
Choosing a resolutionCamera resolution involves two quantities, data resolution and spatial resolution. Data resolution describes the number of bits that represent the gray-scale value from each pixel. For monochrome digital cameras, resolutions range from 8 bits (256 values) up to 16 bits (65,536 values) per pixel. Most color cameras provide "24-bit" resolution, which equates to 8-bit resolution for each sensed color—green, red, and blue.
When choosing a camera, carefully consider how many gray-scale steps you need. Machine-vision systems commonly require only 8 bits or 1 byte per pixel. (Color cameras offer similar steps for each color, from full color to black for each sensed color. And a 24-bit color camera would produce 3 bytes per pixel, 1 byte per color.) If you need a finer resolution of light intensities, look at digital cameras that offer 10- or 12-bit data. But going beyond 8 bits immediately doubles the quantity of data (2 bytes per pixel) and halves the image-transfer rate. Higher data resolutions, also called "bit depths," can increase system complexity because of the increased data processing required to handle the higher bit depths.
A camera's spatial resolution involves two related measurements: the field of view and the minimum feature dimension. Field of view specifies the overall dimension of an image, say 20x15 mm. The minimum feature dimension specifies the smallest feature—a mark or component, for example—that the software or an operator must detect in an image.
Suppose an inspection system must verify that a handler has properly loaded a tray with 24 circular components. The tray measures 400x250 mm and contains the components in a 4x6 matrix. Each component measures 25 mm in diameter. In this example, the camera must resolve the smallest feature, the component.
When a camera acquires an image of a dark tray, the components appear as light circles bounded by two edges with a "span" between them. The edges occur where the image changes from dark to light. The distance across the component (through the center) represents the span.
To select a camera with the proper resolution, first determine the needed minimum feature resolution—or the number of pixels that will adequately represent each edge and the component. In this application, you would allocate three pixels for each edge and four pixels for the span. Thus, 10 pixels (3 + 4 + 3) should define a 25-mm component, and each pixel will represent 2.5 mm. (Equipment and software vendors can help you determine the minimum feature resolution needed for a product.)
 |
| Figure 3. A minimum resolution of 2.5 mm/pixel provides an image in which an operator or software can detect the presence or absence of objects |
Cameras with such a low resolution are not common. More than likely, you would specify a standard camera with a higher resolution, say 640x480 pixels. At that resolution, an image will provide additional information such as the component marking, as shown in Figure 4.
 |
| Figure 4. A resolution of about 0.7 mm/pixel clarifies component outlines and allows identification of component markings. |
So, how can you inspect each SMT component as well as the entire PCB? Several alternatives make sense:
- divide the PCB into areas that several cameras can inspect,
- inspect sections of the PCB with one camera and then move the camera or the PCB to cover the entire board, or
- pass the PCB past a line-scan camera with the required pixel/mm resolution and then use software to build up an image.
In the last case, the line-scan camera must meet the dimensional-resolution along one axis (y). The host computer handles resolution along the other axis (x). The computer monitors the PCB's speed and triggers the line-scan camera to acquire the needed number of image lines per unit distance.
More pixels, slower acquisitionResolution comes at some cost: Adding pixels can reduce image-acquisition speeds. A 200x100-pixel camera produces 20,000 bytes per frame, so at a camera-to-PC data-transmission rate of 100 Mbytes/s, the camera can acquire a maximum of 5000 frames/s. But a 640x480-pixel camera creates 307 kbytes/frame, so at the same transmission rate, the maximum image rate slows to 325 frames/s.
Many cameras, though, operate at a fixed frame rate between 30 and 60 frames/s, which equates to 33 or 16 ms, respectively, for each frame acquisition. But the high transmission rate (325 frames/s for the 640x480-pixel camera) requires only 3 ms of the frame-acquisition period. So, a host PC can use the remaining time between bursts of image data—from 30 to 13 ms—for processing tasks.
Can't you sit still?When you evaluate digital cameras, the motion of objects undergoing inspection will enter into the decision about which camera to choose. The previous example assumed the tray of components did not move during image acquisition. If a device or product moves slowly down a production line, you need not stop the line to capture an image. A short exposure time may suffice to produce a useful image.
Generally, image acquisition should occur in less time than it takes for the object to move by one pixel. Thus, to stop the "action," you control the camera's exposure time, not the product's motion. Say an object moves 1 cm/s and a camera must provide a 1-pixel/mm resolution:
s/exposure =
s/exposure = 0.1
To properly acquire an image in this scenario, the camera cannot operate with more than a 0.1-s exposure period. Some blur still occurs because the product will move by 1 pixel, or 1 mm. Reducing exposure times proportionally decreases blur in images, but shorter exposures demand more light. If an object moves at 1 cm/s and inspection requires a resolution of 1 pixel/micron, the maximum exposure comes to 1x10–4 s. For such short exposures, you'll need strobe lights to sufficiently illuminate the object.
If an application requires a rapid sequence of images, keep in mind that some cameras cannot acquire a new image while they transmit the current image. Even if a camera can acquire an image in only 1 ms, it still may need a "dead time" of 16.6 ms to transmit the image. In this case, the camera has an upper limit of 60 frames/s. Sophisticated cameras may provide a buffer that allows for rapid acquisition of images and then a slower transmission of data.
Choosing the right camera can seem like a daunting task because manufacturers offer so many options for lenses, mounting arrangements, pixel counts, and so on. Limit your selection criteria to basic needs and remember that a camera has only one job: to acquire clear images from which operators and software can extract useful information. With your goals in mind, and the information in this article, camera manufacturers can help you narrow a spectrum of choices. T
| Author Information |
| Chris Brais works as the applications manager and is responsible for product integration and support for the Intelligent Products Division of Coreco Imaging. He graduated with a BS from Boston College and has more than 15 years of experience in imaging, semiconductor, and test and measurement technologies. cbrais@corecoimaging.com. |
| Acknowledgement | ||
| Thanks go to Bill Libiec, support specialist at Dalsa (Waterloo, ON, Canada), for his contributions to this article. | ||
