Geo-Modeling

Major improvements for the existing framework for the generation of orthoimages from 3D laser scans and associated images were made. Namely a memory management system was implemented in order to be able do work with large amounts of input data. The generation of textures was improved and a unified import/export file format was introduced. Several tools for the analysis and conversion of 3D laser data were developed (e.g. robust plane extraction, data conversion to other 3D formats etc.) The MetropoGIS system for image based modelling was improved by implementing a framework for the orientation of images that were captured at widely separated viewpoints. This wide baseline image orientation system is based on scale invariant feature detection/matching and the five point algorithm for estimating the relative image orientation. Additionally a method for extracting robust regions (MSERs maximally stable extremal regions) was implemented. Within the city modelling framework a method for rectifying images of building facades based on extracted vanishing points was developed. On the scientific frontier two feature based methods for image matching and image modelling have been developed. Zwickels - new affine invariant region descriptor for wide baseline image matching was introduced (TR-VRVis-2004-051). The descriptor is based on the region enclosed by intersecting line segments and is especially well suited for matching image pairs of man made structures, where lines are the most prominent features. The modelling approach is called "Sparse 3D Reconstruction by Edgel Sweeping" (TR-VRVis-2004-052) and allows an efficient 3D reconstruction based on edge features that are extracted from the images.