Image Acquisition: Shaping the Future
by Yvon Bouchard, OEM Applications Manager - Teledyne DALSA Posted 04/27/2005
Machine vision applications constantly demand more from the image acquisition process: more bandwidth, more information, higher resolution, and faster performance. In addition, there is pressure for smaller packaging and lower cost. These forces continue to accelerate, with the industry undertaking new initiatives to reconcile the conflicting requirements.
Decreasing Size, Increasing Speed
Camera manufacturers face a dilemma commonly found throughout the high tech industries, that is users want smaller and lighter cameras for vision tasks while also expecting additional features and capabilities. In many cases, such as robotic welding, new applications simply cannot accommodate large, heavy cameras. The speed at which a robotic head moves, for instance, up to 300 times per second when bonding a die to a carrier, makes it physically impossible. As a result, cameras are shrinking as real estate becomes more restricted. One leading camera vendor has downsized its product line from a diameter of four inches to one.
The challenge for image acquisition is that the smaller cameras become the less likely they are to have space available for more sophisticated input functions. Additional capability then falls to the frame grabber, which can include such enhanced functions as preprocessing calculations and sophisticated image analysis, including FFTs and run length encoding.
Frame grabbers must also be capable of handling ever faster manufacturing speeds. As frame rates move up from 30 to 60, 120, even 240 frames per second, the PCI bus itself becomes the bottleneck. Therefore, designs that incorporate onboard frame buffer memory and dual ported architectures will become the norm, able to support line scan rates of 100 MB/sec, 200 MB/sec, 400MB/sec eventually up to 800 MB/sec. As acquisition speeds increase to the 1 GB/sec range, the power of an embedded vision processor will be mandatory.
Standardizing the Interface
In the past, a unique cable has been required to connect every frame grabber and camera. Today there is only one interface between cameras and frame grabbers for bandwidth beyond 100 MB/s. This standard is called Camera Link. Camera Link simplifies the interface between the camera and the frame grabber, decreasing the number of wires between the camera and the board, and simplifying the cabling assembly. The Camera Link interface is designed to meet the needs of high end machine vision systems, with upwards of 100MB of bandwidth. The Camera Link standard also addresses the issue of the size of the connection. Delivering the information captured by high bandwidth cameras has, until now, required complex and bulky cabling assemblies to support 32-bit or 64-bit parallel signals. With the development of high-performance serial links, interfacing has become simpler, with a corresponding reduction in physical size. One objective of Camera Link is to accommodate shrinking packages by keeping camera connectors small.
Camera Link provides benefits for both for manufacturers and customers. With the interface issue out of the way, developers will be able to focus on making the vision solution work, reducing development time, complexity, and cost. Manufacturers and integrators will be able to support more customers and bring them up to speed faster. For customers, the standard will mean the widest choices of products and suppliers.
All Digital
The push to digital technology is inevitable in order to handle increasingly faster speeds and more complex operations. As a result, look for digital interfaces and digital cameras to surpass analog cameras/frame grabbers. Analog to digital conversion will increasingly take place within the camera, rather than passing the analog signals to the frame grabber to perform the conversion. This frees the frame grabber or image processor to handle more complex tasks built into the acquisition, whether preprocessing or implementing software algorithms, whatever is required to speed the process.
Because of their relatively low cost, analog cameras are currently in high demand. Yet the inherent limitations of the technology make them unsuitable in machine vision systems for challenging applications such as semiconductor inspection. In such electrically noisy environments as reticle or LCD inspection, data rates can easily exceed 100 MB/sec. Unfortunately, as the bandwidth increases, the analog signal's sensitivity to noise increases exponentially. By performing the digitization in the camera, the noise sensitivity is significantly diminished and inspection reliability is improved.
An Issue of Color
Look for a move toward color digital vision in the future as the demand for more information from our manufacturing processes continues. Many applications that were handled in monochrome will switch to full color. This move will affect many applications that were mainly in the monochrome domain. Parts inspections, for instance, are often only concerned about sizes of objects; now the color of the object must be inspected too.
In support of the search for low cost color solutions, DALSA Corporation has developed a set of algorithms that allows color images to be processed in realtime from inexpensive Bayer filter-based cameras. Using its Advanced Pixel Processor engine, realtime color decoding can be performed at the full input rate of 30 frames per second or higher, enabling live display of color images without burdening the host PC. This technique is about one quarter the cost of solutions using traditional full color cameras, with similar performance.
Overall, image acquisition will continue to focus on getting more information out of the signal and driving more bandwidth through smaller interfaces. While frame grabbers must continue to acquire images and transfer them out of member, their links to the rest of the world will be simpler. In the end, the real winner is the customer, who will be able to get the best tool for the job as performance demands continue to grow.
