Uma proposta experimental de reconstrução 3D a partir de imagens ortogonais

Abstract

Advances in the last decade in the field of computer vision have brought great benefits to virtual reality applications, with photogrammetry as a major driver. This technique became popular after the emergence of UAVs, adding improvements to areas such as civil engineering, agronomy, and geoscience, making it possible to map a specific region with aerial missions. In this context, this work presents a proposal to perform 3D reconstruction real scale scenarios overflown by a UAV. For this, an approach in C ++ was developed capable of performing this reconstruction for a set of unordered images. The Struct From Motion methodology was used to perform the calculation of the odometer of the cameras at the time of image capture. Finally, the Multi-View Stereo technique is applied to obtain a dense point cloud of scenery. As innovative methodologies, the GPS coordinates of each image are used to obtain a real scale reconstruction and an adjustment during propagation to improve performance in regions with little color variation. The results obtained were satisfactory, where the reconstruction using aerial images presented a defined result and with measures similar to those found on Google Maps, obtaining a maximum error of 1,59%.