Vision Show Presentations Address Gigabit Ethernet and Line-Scan Cameras
Rick Nelson, Chief Editor -- Test & Measurement World, 10/19/2004 7:12:00 AM
| See our complete coverage of Vision 2004: Vision 2004 Highlights |
Mehl of Basler predicted a bright future for color line cameras, with resolutions extending beyond 8k pixels and rates moving well beyond 160 megapixels per second. Color, he said, is important because color techniques can locate defects that gray-scale approaches miss, and he suggested color can minimize operator errors. He described two color line-scan approaches—prism and trilinear—and discussed the pros and cons of each.
Prism cameras, he said, employ three sensors—one each for red, green, and blue—and require a prism to split the incident light into its three components. Trilinear cameras, he explained, use a single, 3-line sensor, but each line has a different field of view that must be compensated for via spatial correction performed in a camera or frame grabber. In addition, a conveyer used with a trilinear camera must incorporate a spatial-based triggering mechanism to prevent the haloing that could accompany conveyer speed deviations.
Then he addressed the question, "Which is better?" The answer, he said, is "It depends!" Trilinear units, he explained, are suitable for price-sensitive applications as well as those requiring cameras having low weight and compact size. Trilinear cameras can be eight to ten times more compact than prism cameras and can require expensive, custom lenses.
However, prism cameras, he said, support higher line rates (greater than 10 kHz) and are adept at applications involving falling targets or targets that otherwise can’t be definitively aligned with the camera.
All and all, he expects the market for line-scan cameras to increase 15% per year, with trilinear versions grabbing the largest market share. Pleora’s George Chamberlain described the applicability of Gigabit Ethernet to vision applications, describing potential problems and solutions. High-speed Ethernet, he said, can provide a high-speed direct link from camera to PC, outpacing rates available with interfaces such as Camera Link. In addition to providing a single-camera to single-PC interface, Ethernet enables other architectures, including one camera to multiple PCs (in which each PC might apply a different algorithm to camera data) or multiple cameras to one PC (a configuration that might be used in security applications).
Ethernet, however, introduces difficulties for machine-vision applications, especially with respect to real-time performance. For imaging applications, frames must be available at a constant and consistent rate, a requirement that the latency inherent in Ethernet can compromise. Latency, Chamberlain explained, stems from packet formation at the camera as well as data transfer to memory on the PC host, especially if you rely on the Windows protocol stack. Another problem with Ethernet is the way it handles lost or out-of-order packets—each of these conditions imposes delays that exacerbate latency, which can measure in milliseconds. Ethernet workarounds such as UDP can minimize the latency penalties but at the expense of failing to detect lost packets.
To overcome such problems, Chamberlain proposed Pleora products such as an FPGA-based IP engine that resides in or near a camera; that engine, he said, eliminates latency caused by packet formation. He also proposed employing a Pleora IP device driver at the PC end of the link to speed data transfer to memory without CPU overhead. The result, he said, is total latency of less than 300 microseconds, which he noted is up to 80 times less than provided by other approaches.
For more information from George Chamberlain, see the November Test & Measurement World Machine Vision & Inspection Test Report, online November 1 at www.tmworld.com.
