Surface reconstruction from stereo images preserving discontinuities
Naokazu Yokoya · NAIST Digital Library (Nara Institute of Science and Technology) · 1994
This paper describes a method of surface reconstruction from binocular stereo images. The problem addressed is formulated as the minimization of an energy functional that integrates a similarity term and a smoothness term realizing a controlled-continuity constraint. The minimization reduces to solving a large system of non-linear equations which is a discretized version of Euler-Lagrange equation. A disparity map, which is a real-valued function of the image coordinates, is obtained by solving the system of equations through scale hierarchy. Initially assuming a flat (constant disparity) surface, the proposed method first finds a local minimum of energy functional at a coarse scale and then proceeds to finer scales by using the previous result as an initial estimate at the current scale. At the finest scale, discontinuities and occluded areas are detected from a computed disparity surface. The proposed method can compute a dense disparity map directly from stereo images preserving discontinuities.