Pintojen rekonstruointi 3D-pistepilvien avulla
Pekka Ekman · Aaltodoc (Aalto University) · 2017
Automatic reconstruction of 3D polygon scenes from a set of photographs has many practical uses. We have implemented and tested two alternative processes for such reconstruction. The processes consist of separate pieces of software, which together form a pipeline. We used three existing applications: COLMAP, RAPter, and ManhattanModeler. We also created a texture generator for including color information into the reconstructed scenes. We constructed two alternative pipelines from these four applications. First, a sparse and a dense point cloud were generated from the set of photographs using COLMAP. Then, a polygon representation of the scene was generated using RAPter in the first alternative pipeline and ManhattanModeler in the second. Finally, textures were generated and added to the polygon scene using our texture generator, which generates textures from the color information in the dense point cloud. We tested and evaluated these pipelines using a set of photographs taken in Aalto University Computer Science building library with mostly acceptable results. There were two notable imperfections: RAPter generated significantly oversized polygons, and ManhattanModeler failed in its last reconstruction step, which we had to perform manually. Texture quality was reasonably high in many places. However, the texture generation process was highly sensitive to imperfections in the polygon scene, resulting in poor texture quality in some places.