Connection machine stereomatching

Abstract

This paper describes a parallel real-time stereomatching algorithm and its implementation on the Connection Machine™ computer, a new massively parallel computing system. The main features of the algorithm are 1) real-time performance, 2) the full exploitation of the ordering constraint, 3) a representation that easily maps onto a parallel computer architecture, and 4) the ability to efficiently use a variety of matching primitives. Some results, including timings, are shown for both real and synthetic data. Also discussed are the use of color information and some subtle variations of the basic algorithm.