How do you think the new GigE standards will influence the machine vision industry?
Respond or ask your question now!
By Lee J. Nelson
Dr. Heric: Data processing is vastly different. The old “minicomputers” are distant memories. It’s difficult to imagine those monsters relative to today’s desktop work environment. The vast array of data products, also, is a true advancement. Two decades ago, if one wanted a scene, choices were limited to what early adopters had archived. Now we have floating scenes, ready-processed data and blends of data types.
Dr. Levine: There have been several interesting developments in the past twenty years, often as a result of the availability of very fast, inexpensive computers. Model-based recognition largely has been replaced by appearance-based methods in which the visual appearance of objects—not their features—are used to characterize them. This led to a significant increase in research on the modeling of illumination effects, associated with the image-acquisition process. The same holds true for object detection and tracking in video imagery. As a consequence, research in facial recognition, video processing and real-time object tracking has become quite practical, although still limited in scope. Another noteworthy outcome has been the convergence of computer and biological vision, with each side feeding on the other, to mutual advantage.
Mr. Schatz: We’ve matured from a venture capital funded battlefield into a multi-billion dollar industry with a well developed infrastructure of experienced suppliers, integrators, and most importantly, customers that know well how to apply our products. But, we’ve also become a highly fragmented business with distinct niches defined by industry and applications. From a hardware perspective, we all started out in the 1980s with expansive embedded systems—moved to PC-based platforms in the 1990s—and back to embedded systems based on DSPs and microprocessors in the post-2000 era. Now, however, they are compact and inexpensive versus clunky and expensive. Software evolved from total reliance on binary SRI [Stanford Research Institute] algorithms to sophisticated shape matching and analysis techniques. Ironically, we now are seeing some new machine vision players entering the market with sensors that basically take legacy SRI-style algorithms and run them on contemporary hardware.
AI: What was the most unanticipated development?
Mr. Biegel: The forecasted death of existing imaging modalities was premature. For the most part, every novel modality has supplemented—rather than displaced—existing diagnostic imaging methods. Nuclear medicine is a good example where new applications have evolved through synergistic exploitation rather than replacement.
Mr. Gibson: Two key items came from nowhere to improve the imaging industry. Plastic lenses made possible single-use film cameras, low-cost and mass produced digital cameras as well as personal imagers such a cellphone cameras. The other is pervasive acceptance of the Microsoft operating system environment which has demystified most image algorithm development processes.