- This article is filed under:
- » View All
New Machine Vision Cameras Fill the Bandwidth Pipe
by Winn Hardin, Contributing Editor - AIA Posted 08/19/2003
Cameras, like machine vision systems in general, have improved directly as a result of faster microprocessor speeds. Faster microprocessors and improved data communication schemes such as CameraLink and IEEE 1394 mean that PCs and vision controllers can handle more information. These bandwidth improvements are allowing camera manufacturers to answer the calls of their customers for improved read out speeds, higher resolution and additional functionality.
Increasing the data pipe
Standardized data interfaces represent one of the most active areas of machine vision camera design work. According to Dr. Joachim Linkemann, product manager at Basler Vision Components (Ahrensburg, Germany),’‘Customers are asking first for standardized interfaces such as CameraLink and Firewire [IEEE 1394]. Firewire has a limitation of 400 Mbps, which in most cases is absolutely sufficient. We are developing CameraLink cameras while including an option of an interface converter so that customers are able to use their old low voltage differential signal equipment.’‘
Basler is working with CameraLink standards organizations to enhance the full configuration to handle more than 8 taps per a camera while still delivering up to 500 MBps in full configuration. Basler’s latest release is a 10 tap, 500 frames per second 1.3 megapixel camera capable of filling a CD-ROM with image data in one second. New PCI buses, such as the PCI-X capable of burst transfers up to 1 GBps, are also helping to make high resolution, high speed cameras a reality, although the cost of these systems will need to continue to drop before the machine vision industry can take full advantage. Today, a PC using two 1394b I/O cards would be at the PCI bus’ bandwidth limit.
Linkemann said that Basler is also using the less expensive Firewire standard to improve the cost effectiveness and appeal of digital cameras. ‘‘If you take the whole system into account – camera, cable, frame grabber – a digital system using Firewire can be competitive to a high-end, analog camera with all the extra components.’‘
Basler is also seeing a significant increase in the demand for color machine vision cameras. ‘‘Currently, the ratio is 2 monochrome cameras to 1 color, but we think the ratio will be 1 to 1 within the next two years,’‘ Linkemann commented.
While color cameras may only represent 20 percent of total industrial camera sales at Sony Electronics Inc.'s Visual Imaging Products division (Park Ridge, NJ), one of the largest providers of machine vision cameras, digital cameras are another affair. ‘‘Last year, our revenues from the sale of digital cameras were higher than analog cameras, but we still sold more analog units. While analog cameras will have legs for many years to come, within in the next two years, expect to sell more digital cameras into machine vision than analog,’‘ predicted Jerry Fife, senior product manager at Sony Electronics for industrial and scientific cameras.
Flexibility, speed and size are driving Sony’s new machine vision cameras. Sony has released two new 1.25-in cube cameras that weigh only 50g for vision guidance applications, while doubling the speed and increasing the resolution (640x480 at 60 f/s and 1024x768 at 30 f/s). On several camera versions that offer increased resolution, Sony gives the end user the flexibility of higher frame rates at lower resolutions through on-camera 1x2 pixel or 2x2 pixel ‘binning,’ as well as a ‘quick scan mode’ where the operator can set a region of interest within the full array. ‘‘One of the keys to using a shared resource like 1394 is bandwidth management,’‘ Fife explained, ‘‘particularly if you’re going to have multiple cameras transmitting images simultaneously. Networking multiple cameras and other devices within a system is the one of the great advantages of 1394 over direct point-to-point connects of CameraLink.’‘
Camera manufacturers fill the data pipe as fast as new communications protocols can expand to meet the need. Dalsa (Waterloo, Ontario, Canada) newest line scan cameras are excellent examples of the bandwidth-hungry machine vision image acquisition market. Line scan cameras offer high resolution for a fraction of the price of area scan cameras. ‘‘It’s more economical to have an 8kx1 line scan camera than an 8k x 8k area scan camera, or a 4K or 2K line scan camera’‘ explained Dalsa’s line scan product manager, Mark Butler.
According to Butler, the performance driven line scan camera market is addressing leading growth sectors including printed circuit board and flat panel inspection -- areas that need high resolution but are concerned about costs.
‘‘More speed and higher resolution – those are the trends we’re seeing. In the past it’s been responsivity, dynamic range and speed, and we found that responsivity and speed were the two main categories that customers needed. Now, the cameras have improved such that all three are critical.’‘ Dalsa is working with its foundry to use a variety of methods to improve line scan camera performance including improving the sensor design to optimize the charge conversion efficiency and quantum efficiency.
The need for speed and resolution are leading to multi-tap line scan cameras. Dalsa has recently improved the speed of its multi tap cameras to 40 MHz. When the speed of the system impinges on lighting budgets or systems are light-starved, Dalsa offers time delay and integration (TDI) cameras, an ultra-high sensitivity variant of linescan imaging, that are essentially several line scan cameras linked together. The charge from one pixel #1 in the first line array is transferred to pixel #1 in the second line array as the object under inspection moves along below the camera. Each line array adds its charge to the next until sufficient light is acquired for a sharp image.
‘‘Synchronization is the number one issue with TDI cameras. The good news is that the physical inertia of many of today’s industrial processes restrain non-ideal object motion thereby reducing synchronization issues’‘
When color is important, Basler synchronizes three line scan arrays together to create an RGB line scan camera. ‘‘This is a huge advantage over older color line scan cameras that used prisms inside the camera,’‘ Linkemann said. ‘‘Costs are reduced and its is easy to use, but there are a few conditions that have to be followed to have a real RGB image: you need an encoder that provides a spatial trigger not a time dependent trigger.’‘
Cameras may soon reach a physical limit similar to the challenges posed to the microprocessor world. According to Basler’s Linkemann, as pixel sizes continue to shrink, lenses become more of a factor. ‘‘The pixel is going down to 3.5 or 5 microns. From the wafer manufacturer side, this makes sense because you can have more CCDs on a single wafer. But the physical resolution is limited by the optics. It has to be a very good and expensive optic. Currently, a 3-megapixel camera is close to the limit and 5 megapixel cameras are already over designed. You need to have better optics or larger die sizes.’‘
As illustrated here, convergence can be a good and bad thing. Improvements to semiconductors are improving machine vision systems, but also shifting system constraints from the processors to the cameras and back to the processors as PCI buses become overloaded with image data. However, performance is measured in many ways. Digital cameras with more on-board processing will also enable new levels of camera control including gamma correction, shading correction, color interpolation among others.
There are currently no comments for this article.
Leave a Comment:
All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.