Three-dimensional surface reconstruction using computational stereopsis

Abstract

A computer process for reconstructing 3D objects using two spatially separated photographic images is introduced and implemented. Normalized cross-correlation is used to locate feature points which are common to both 2D images. The displacements of these feature points are used to generate disparity maps, and disparity values can be used to compute x,y,z locations. Since disparity maps are typically noisy and not fully populated with known disparity points, a mathematical curve fitting is necessary to fill in missing points and smooth the data. Adaptive smoothing algorithms have been investigated and used to develop a method for refining the disparity map and reconstructing 3D surfaces.