Parallel stereo matching of SPOT satellite images

Most recently, an area based region growing stereo matcher [Otto & Chau 1989] using an adaptive least-squares correlation was developed by the Alvey MMI-137 project. This algorithm (coded in C) is being used as the second step in a completely automated system for extracting 3D co-ordinates of terrestrial and industrial surfaces from stereo images, the first step is being the extraction of some conjugate points `seed points´ in the two stereo images [Allison, Zemerly and Muller 1991]. This stereo matcher requires about six days (cpu) for a full sized 6000×6000 pixel SPOT pair processed serially on Sun sparcstations running SunOS. The objective of this work is to match these pairs in about two hours using the PARSYS SN1000 (48 T800+4MB-RAM) Supernode based transputer machine