
Since system speed is an all-important constraint, specialized image processing/computer vision hardware is used to implement the aforementioned segmentation algorithms.
The vision industry is following two design strategies, one in which hardware is tailored for specific vision processing tasks, and the other in which general DSP chipset is utilized to provide more generic capabilities. The former strategy provides for greater performance if one's algorithm can be decomposed into a series of subtasks implementable by the hardware. The latter strategy compromises speed for computational flexibility. The Aries vision system falls within the first category.
Aries #1 employs a modular vision system, manufactured by
Imaging Technology Inc. (ITI), with full scale pipeline processing
capabilities. The system uses two ITI IMA
memory managers, each with four Mbytes of
reconfigurable memory. These cards are designed to carry submodules
that perform different image processing operations. One of its
design features is a multi-input-multi-output cross-port switch
that allows the reconfiguration of the pipeline for different
operations. The current system uses two ITI IMA
memory managers, each with four Mbytes of
reconfigurable memory.