View Planning Using Geospatial Information for 3d Reconstruction with Unmanned Aerial Vehicles

This paper describes a method that uses geospatial information to propose view planning for structure from motion-multiview stereo (SfM-MVS) with unmanned aerial vehicles (UAVs). In this study, view planning using scene geometry was extended to 3D space based on previous studies. However, this method requires many viewpoint candidates, which requires a large amount of computation. Therefore, geospatial information is used to reduce the amount of computation, for example, by deleting viewpoint candidates that cannot be passed through by UAVs. Image capturing was performed using a 3D model generated in OpenStreetMap. To verify the 3D reconstruction from the planned viewpoints, we obtained a 3D model of the simulation from the planned viewpoints, and this set of images was used for SfM-MVS. Consequently, 3D models could be reconstructed; however, various challenges were encountered.