A parallel vision system for object recognition and location based on cooperative depth and intensity processing is described. The parallel algorithm for intensity data processing is based on generation of hypothesised matches between line junctions in the image and space curve intersections in the model. These hypotheses lead to back-projection of the model and verification of promising hypotheses. The parallel algorithm for depth data processing is based on a tree search algorithm constrained by pairwise geometry between primitives. As each algorithm proceeds, partial results are interchanged to direct the other concurrent process to a more promising or more viable solution. The architecture has been implemented and evaluated on a multi-transputer machine, and is illustrated by several examples of pose definition of a test object. © 1997 by John Wiley & Sons, Ltd.
|Number of pages||22|
|Journal||Concurrency: Practice and Experience|
|Publication status||Published - Feb 1997|