Disparity smoothing algorithm for three-dimensional scene reconstruction

Methods allowing a three-dimensional model of the real world to obtain have been establishing a vivid branch of research in the machine vision area. Among them one very popular is the stereo vision. Stereo vision is a method allowing to obtain a good realistic model of a real scene. However, the central step in stereo vision reconstruction is the image matching. All algorithms which solve the matching problems suffer from holes and unmatched areas in their result so called disparity map. These wrongly matched regions make almost impossible visually satisfactory reconstruction. In this paper an algorithm for smoothing disparity map is proposed. A full algorithm for reconstruction with the disparity smoothing step is sketched. The usefulness of disparity smoothing is confirmed by its application to the reconstruction issue of real stereo pairs.