An integrated stereo algorithm based on coarse-to-fine features and intensity values

Abstract

Depth perception is a major problem in computer vision. Stereo vision is one of the most useful techniques for depth perception because it gives quantitative depth measurements. Computational stereo is defined as the recovery of three-dimensional characteristics of a scene from two images taken from two points of view. In this work, a stereo algorithm is presented. Unlike most of the previous stereo algorithms, which are classified as either feature-based or intensity-based, this algorithm makes use of features as well as intensities. Features are extracted using Laplacian-of-Guassian filters of different sizes. The algorithm operates in a coarse-to-fine manner with the lowest level being the image itself to account for intensity matching. The algorithm was tested on a variety of images and gave good results in terms of accuracy as well as short running time. Keywords: Stereo vision, Depth perception.