By giving machines some form of vision, manufacturers have gained a powerful quality control tool. Machine vision systems can capture images and measure a product's dimensions, position and color, component locations, or other key characteristics, providing rapid "pass/fail" judgments even when unattended.
All machine vision systems include a camera, a computer, and software for capturing and analyzing images. The selected system components must meet the needs of the specific application. Because the image sensor determines the speed and resolution of the imaging system, the selection of the correct image sensor has a critical impact on the success of vision applications. The different image sensor architectures used in machine vision, particularly those related to electronics manufacturing, will be discussed below.
Figure 1 shows an image sensor that converts light into electrical charge and reads out the charge signals in a certain order, thereby reconstructing the image information.
Figure 2 Line scan structure
In Figure 3, the pixels of the full-frame architecture both convert light into electrical charge and function as readout areas circuits.
Various architectures of CCD
All CCD sensors possess a large number of light-sensitive points (pixels) and a readout mechanism. The former converts incident photons into electrical charges, while the latter sequentially transfers the charge from each point to an output amplifier, thus moving it outside the sensor. Various designs have been developed, often emphasizing certain performance aspects while sacrificing others. Therefore, some image sensor architectures are better suited for machine vision, while others are more suitable for a wide range of applications, from astronomy to amateur and professional digital photography. Machine vision systems employ various sensor types, including linear arrays, line-to-line transfer, full-frame, and frame-transmission.
Linear image sensor
A linear image sensor (line-by-line scanning) comprises one or more linear arrays of pixels. Each array is coupled to at least one readout device and amplifier. Linear image sensors are suitable for machine vision applications that require imaging of continuously manufactured products, such as PC boards on conveyor belts, future printed plastic circuit boards, and other thin, rolled products, such as magazines, printed fabrics, and/or banknotes (see Figure 2). In summary, linear sensors have a simple overall structure and are suitable for imaging flat, fast-moving objects, but they often cannot compete with area sensors in applications requiring the capture of 3D object images.
Full-frame sensor
Full-frame sensors combine photosensitive sensing with readout. Since there is no separate storage area, an external shutter (or synchronized strobe illumination) is required to prevent incident light from illuminating the pixels before any charge transfer occurs (see Figure 3). Without a shutter or strobe, the image will exhibit a trailing or smudged effect.
In the early days of machine vision (mid-1980s), full-frame area sensors were used because they were the only available option with sufficient resolution for the application. If an application required the resolution offered by a 1024×1024 pixel sensor (such as Kodak's KAF-1400 sensor), then a full-frame sensor was the only choice.
In summary, full-frame sensors have the simplest architecture among all area-type sensors, and offer the highest resolution and density of light-sensitive areas (the latter referring to their highest fill factor). They also provide high full-well capacity, low noise, and a large dynamic range. However, they require a mechanical shutter.
Frame transmission image sensor
A frame-transmitting image sensor is similar to a full-frame imager. However, it employs a second planar array that provides light shielding and serves as an image storage area (see Figure 4). This structure does not require a mechanical shutter, thus achieving a higher frame rate than full-frame sensors because they can acquire one image while transmitting another. However, since integration still occurs during image transfer to the storage area, image trailing artifacts can occur, impacting performance. Because implementing this architecture requires doubling the integrated circuit area, frame-transmitting CCDs generally have lower resolution and are more expensive than full-frame CCDs.
In summary, frame transmission sensors offer higher fill factor, higher full-well capacity, lower noise, wider dynamic range, electronic shutter, and better frame rate. Their main drawbacks are significant image smudges at very short exposure times and higher manufacturing costs.
Figure 4 shows a frame transfer architecture that collects light from one array (top), then transfers the charge to another array that achieves optical shielding, and then reads it out.
Figure 5 shows an inter-pixel sensor architecture that transfers the charge that generates light to a shielded area adjacent to the pixel. The sensor loses its image-sensitive area, but its speed is increased.
Interline Transfer Sensor
In line image sensors, light sensing and readout functions are also separate. Each pixel is surrounded by a light-shielded vertical CCD (VCCD) that can transfer charge (see Figure 5). This allows the line sensor to capture one frame while moving the previous one away, thus enabling a built-in electronic shutter capability.
Line-to-line sensors were developed later than full-frame and frame-transmission sensors. With the maturation of line-to-line technology, it has been able to provide the higher resolution and frame rates required for machine vision. Today, line-to-line sensors are the most commonly used sensor form for 3D object imaging, with applications including verifying the correct placement of components on circuit boards and inspecting packaged ICs to ensure that leads are not bent.
In summary, inter-line transfer sensors offer resolutions from VGA to megapixels, along with high full-well capacity. They also feature low noise, wide dynamic range, electronic shutter, high frame rates, and low smudges, enabling short exposure times.
Conclusion
In summary, users desire faster frame rates (to keep up with fast-moving objects), higher quantum efficiency (to provide more images in low light and/or with shorter imaging times), and greater dynamic range (so that relative detail can be seen in brighter or darker parts of the image). Electronic shutter, progressive scan readout, and high sensitivity are all key parameters to consider when determining which sensor is best suited for machine vision applications. It should be remembered that it is the matching of this entire set of parameters that makes a particular sensor the optimal choice for an application.
