A Methodological Proposal for Improvement of Digital Surface Models Generated by Automatic Stereo Matching of Convergent Image Networks

A Digital Surface Model (DSM) generated by automatic stereo matching of convergent image networks includes a great number of 3D-points that come from diverse combinations of stereo-pairs. This points can have a very different accuracy and reliability, in addition to a great and nondesirable spatial redundancy. We analized different methods to generate several synthesis DSM whose points are the result of a statistic process. The aim is to obtain a regular mesh of points whose Z coordinates are estimated from the whole set of data over the interpolating function. We choose the best of them using a random mesh of points whose coordinates were measuredby multiple direct intersection.