A special architecture for dynamic programming

Abstract

In robotics, pattern recognition systems often have to take decisions and provide results to higher level systems. They can do this simply by refering to a dictionary of patterns in a well defined and limited universe. This paper describes a multiprocessor system which makes in real time a matching by using dynamic programming between a segmentation of reference and an unknown segmentation, in order to provide a likeness coefficient. Application in pattern recognition (speech, vision) but also in decision (dynamic programming in local area ie edge detection) or even in general parallel computer.