Using surface model to correct and fit disparity data in stereo vision

Abstract

A method of stereo vision adapted to 3D reconstruction of urban scenes is presented. A generic geometric model for objects in the scene is assumed. Uniform intensity regions in the images are hypothesized to correspond to single surfaces in the scene. The hypothesis is either verified or refuted by testing whether matches in each region fit the model or not. Disparity errors are detected using a disparity gradient limit before interpolation. The method is tested on real images, and results are given.<>