Constrained sliding window correspondence algorithm for fast stereo vision

This paper proposes and evaluates a novel algorithm for local correspondence matching in stereo vision, named Constrained Sliding Window (CSW). Conventional local algorithms compute the disparity map based on intensity values of pixels within a window for left and right images. Local algorithms are considered to be faster than global methods and capable of implementing applications which require prompt responses. Nevertheless, local algorithms exhibit the critical disadvantage of having a fixed search space, resulting in repetitive scanning. The main objective of this paper is to dynamically constrain the search space to reduce unnecessary scanning and hence reduce the processing time. The proposed CSW algorithm was proven to significantly reduce processing time by up to 45% compared to unconstrained algorithms. The proposed algorithm was evaluated experimentally using Tsukuba image pair and stereo data set from the Middlebury database and was compared against traditional algorithms.