Prof. Dr. Bernd Jähne, University of Heidelberg
Two trends are at work as automated visual inspection systems continue their rapid progress. First, it is demanded that smaller features be inspected on larger targets. Representative applications include the scrutiny of the manufacturing processes for liquid crystal displays and of electronic circuits. Second, visual inspection systems are asked to keep up with the increasing throughput of production lines.
Both call for cameras with faster frame rates and higher spatial resolution. In addition, complex inspection tasks require more than one feature to be analyzed. The best way to do this involves a multisensor camera. The most common type is a color camera that images a scene in three colors: red, green and blue. But there are many other possibilities. One interesting variant employs different types of illumination with each color channel.
In fact, what is critical is not the frame rate per se but the effective pixel clock rate, which is the product of the frame rate and the number of active pixels. Standard cameras with one output channel reach throughputs of between 10 and 60 megapixels per second. For physical reasons (i.e., the capacity associated with the accumulation of charge), this cannot be overcome without compromising the image quality. Therefore, this pixel rate could be increased only marginally over the past two decades.
Higher pixel throughput requires output over many channels in parallel. With CCD technology, this was a complicated procedure and thus was restricted to a narrow sector of expensive high-speed cameras. With CMOS technology, in contrast, it is easy to integrate the additional circuits for parallel output, including all the analog-to-digital converters on a single chip. Therefore, CMOS has opened new doors in the development of inexpensive, high-speed cameras.
A good example of such modern CMOS chips is the MT9M413 sensor from Micron Technology Inc. of Boise, Idaho, which is used in many contemporary high-speed cameras. This chip features 10 digital outputs, 10-bit resolution and a maximum clock rate of 66 MHz, resulting in a pixel rate of 660 megapixels per second. Given this, it is apparent that the limit for high-speed imaging with standard computer equipment is not set by the current digital camera technology.
But there are two other critical factors to consider.
The obvious one is the limited throughput of standard input/output bus systems in PCs. Unfortunately, progress in this field was limited over the past decade. Mainstream PCs still stick with the standard PCI bus system, which has a maximum throughput of 132 MB/s at 33 MHz. The introduction of a higher-speed version that would quadruple the throughput to 533 MB/s was at first not very well received.
Only recently, a new concept for serial data transmission with lower pin counts, the PCI Express bus system, is gaining momentum and can be expected to quickly replace the old PCI bus system. This would finally open a way for higher-speed imaging with standard PC equipment.
The less obvious limitation for high-speed imaging is related to the demand for higher irradiance. The problem is that an image of good quality requires the same exposure, so the required irradiance is inversely proportional to the exposure time. Fortunately, significant improvements in illumination systems are ahead. LEDs already have emerged as the most important illumination source in machine vision systems. Now they are about to change the general illumination market.
Significant progress can be observed in two critical areas of high-power LEDs: The radiant efficiency has increased significantly, and systems for improved heat dissipation have been developed. High-power LEDs with up to 1 W of radiant flux have become available. Thus, much more efficient lighting systems can be expected to be available in the near future to facilitate high-speed imaging.
With falling limitations in pixel throughput, larger-area cameras with the same frame rate as lower-resolution cameras become feasible. This is true both for line- and area-scan models. This results in two significant improvements: A large-area inspection system can be built using fewer cameras, and the resolution of an inspection system can be improved for the better detection of smaller defects. In this way, higher pixel throughput goes hand in hand with higher spatial resolution.
Higher pixel throughput, moreover, makes color imaging with three channels faster and easier. Therefore, another important trend is the merger of fast cameras with flexible LED illumination systems. A target can be imaged with illuminations that differ in color or type, one after the other, using a single camera. In this way, the inspection setup can be optimized to make critical defects more visible and to simplify segmentation and classification.
Consolidation of standards
A word is required concerning the standards for digital cameras. Today, a number of standards are available, including FireWire (IEEE-1394 and -1394b), Camera Link and USB 2. The trouble with these standards is less the sheer number of them than the fact that they are not done well. The Camera Link standard, for example, specifies the physical layer but not the control of the camera (i.e., how to set the exposure time or gain, etc.). With any of them, a user still cannot switch among cameras from different manufacturers without changing or adapting the driver software.
It can only be hoped that the industry is working to the mutual benefit of industry participants and of the users of digital cameras as it develops a new standard. Given the need for high-speed processing, an advanced standard also should include the automatic distribution of a digital stream of image data to different computing nodes for fast parallel processing. This is envisioned in the GigE Vision standard.
Meet the author
Bernd Jähne is affiliated with the University of Heidelberg in Germany; e-mail: firstname.lastname@example.org.