A systematic implementation of image processing algorithms on configurable computing hardware

Configurable computing hardware has many advantages over both general-purpose processors and application specific hardware. However, the difficulty of using this type of hardware has limited its use. An automated system for implementing image Processing applications in configurable hardware, called CHAMPION, is under development at the University of Tennessee. CHAMPION will map applications in the Khoros Cantata graphical programming environment to hardware. A relatively complex automatic target recognition (ATR) application was manually mapped from Cantata to a commercially available configurable computing platform. This manual implementation was done to assist in the development of function libraries and hardware for use in the CHAMPION systems, as well as to develop procedures to perform the application mapping. The mapping techniques used were developed in such a way that they could serve as the basis for the automated system. Many important considerations for the mapping process were identified and included in the mapping algorithms.

The manual mapping was successful, allowing the ATR application to be run on a Wildforce-XL configurable computing board. The successful application implementation validated the basic hardware design and mapping concepts to be used in CHAMPION. Nearly a tenfold performance increase was realized in the hardware implementation and performance bottlenecks were identified which should enable even greater performance improvements to be realized in the automated system. The manual implementation also helped to identify some of the challenges that must be overcome to complete the development of the automated system.