Use of surface reconstruction for machine vision

Abstract

Two cameras with parallel optical axes at the same altitude are often used for stereo vision. In this paper, mathematical relationships between pixel positions and the coordinates of the point viewed are derived for any camera structure. A detailed description of epipolar geometry is included and equations for epipolar lines are derived. The key problem in static stereo method for surface reconstruction is the correspondence problem, the task of matching corresponding pixels in two images. Matching all pixels is necessary for construction of a dense surface. Since it takes an extraordinary amount of time for matching all pixels, construction of polyhedral approximations for surfaces is proposed. An existing algorithm is modified for this purpose.