| The parts inspection industry requires image processing machines operating at speeds unattainable by microprocessor systems. Parallel processors can provide higher speeds, but often the speeds required can only be achieved using custom hardware. Unfortunately, building machines from custom hardware involves long and expensive design cycles. However, this time and expense can be dramatically reduced by building machines from reconfigurable vision systems (RVSs), consisting of hardware modules that can be connected by plugging them into special motherboards. Using such modular hardware reduces system design time and cost by practically eliminating the hardware design and fabrication stages, but writing and testing programs to run on modular hardware networks is still time consuming, and requires detailed knowledge of the hardware.;This thesis presents the concepts behind and implementation of a new design environment, in which users can quickly design high speed image processing machines from a set of modular hardware blocks. Like design tools for multiprocessor systems, this environment allows a user to write and test block diagram algorithms, partition them onto hardware networks, simulate the timing for each module, and generate executables for any programmable hardware. Unlike previous tools, the new environment models custom hardware as well as processors, assists in optimizing processor software, and includes tools to design all aspects of an image processing machine, such as hardware configuration and user interface. It is able to add these capabilities by leveraging structural elements common to image processing programs. The environment, called RVMDES, will be basis for a real tool to design reconfigurable vision machines from the modular hardware developed by the Egypt group at CMU. |
