Illumination-Guided Stereo Correspondence

Abstract

This work implements a method to improve correspondence matching
in stereo vision by using varying illumination intensities from an
external light source. By iteratively increasing the light intensity on
the scene, different parts of the scene become saturated in the left
and right images. These saturated areas are assumed to correspond
to each other, greatly reducing the search space for stereo
correspondence and increasing robustness to erroneous matches.
The stereo camera and light source used in this work is the DUO3D
camera by Code Laboratories. Visually, experimental results show
the resultant point clouds from the proposed method is less noisy
with fewer outliers compared to standard block matching method,
but produces fewer matches.