Faster data-structures for real-time stereo matching

An efficient stereo image matching algorithm, suitable for real-time passive range mapping and navigation, and utilising a novel data structure based on a binary tree, is presented. The data structure, referred to as the 'Edge Priority Tree' or EPT, can be used to reduce the computational complexity of existing schemes which make use of feature points in the images, such as edge strengths, to perform the matching. The EPT can be constructed in O(Nxlog(log(N))) average time for N such values being inserted, sorting them from strongest to weakest in a top-down manner in the tree, while preserving the relative order in which these values were inserted. This facilitates a simple implementation of commonly used constraints, such as the ordering and disparity gradient limit constraints, thus reducing the search space in which matching pixels can be found.