Making research more practical

In preparing each month's issue of Vision Systems Design, the editors examine numerous newswire services, Web sites, publications, and press releases for the latest industry products, technologies, applications, and developments. The magazine's mission is to present useful information that informs its readers of the latest vision and imaging devices and how designers and developers are integrating them to solve today's system problems. In this endeavor, we often come across related university Web sites and academic papers with such esoteric titles as Neural mechanisms for image processing and feature extraction, Insect sight and machine-vision algorithms, and Modeling of geometric objects with the objective of recognition, localization, and interpretation.

During careful examination of these research topics, we are frequently confronted with high-level equations, mathematics, and theories distinctive to the author. Worse, after thesis publication, some researchers often abandon their projects, forego the academic lifestyle, and opt for a position in the telecommunications or semiconductor industry. What's left behind is an idea or method that may not be further developed.

However, some related academic research has proven useful. For example, neural networks have been used for cell classification, modeling insect sight has been used to develop foveal image sensors, and geometric pattern-matching has been used in machine-vision systems. Unfortunately, though, only a few researchers are studying subjects that would dramatically impact the development of future image-processing and machine-vision systems. Many chosen studies are simply too obscure for useful exploitation.

Researching various subjects is necessary to determine whether they hold the promise of practicality. Just as a drug company may spend millions of dollars developing a new medication that is eventually banned by the Food and Drug Administration, vision and imaging researchers often prefer to tackle previously unexplored subjects regardless of eventual usefulness.

Practicality needed

However, just as the goal of drug companies is to develop medicines to control diseases, vision researchers should consult the supplier and user communities to establish a definitive practical goal before beginning a research program. When tied to a specific problem, such as improving the precision of vision-guided robotics, the investigation of vision algorithms for three-dimensional range data acquisition would take on new meaning.

Rather than pursuing a subject solely for obtaining theoretical data and results, vision and imaging research should focus on how the accumulated information could be applied to improve system operation, accuracy, and reliability. This approach would benefit both academia, which could then profitably license the resulting methods and algorithms, and systems designers, who could possibly develop them for practical applications. It need not apply only to vision algorithms; it could be applied to a range of topics, such as neural-network ICs, hybrid SIMD/MIMD architectures, and custom CCD designs.

Once researched and developed, new vision or image-processing hardware and software products would inevitably make system integration, development, and application easier, faster, and cheaper—attributes that rank highly with designers, suppliers, and users. To achieve this, universities and colleges should work together more closely with manufacturers and vendors of imaging equipment. Having high-level researchers involved with imaging product development earlier would spur more novel product development. Similarly, providing researchers with access to the latest manufacturing technologies and marketing data could only result in greater cooperation between researchers and manufacturers.