Exploiting Parallelism in a Monte Carlo Image-Matching Algorithm

摘要

Recovering three-dimensional information from a visual scene requires producingthe disparity map of matched stereo images. The matching approach discussed in this paper involves the Metropolis algorithm, a simulated annealing technique. Although simple in principle, this stochastic optimization technique requires immense computational resources. Adapting the algorithm for parallel computing is discussed and demonstrated for several computer architectures.