Improving 3D mesh quality using multi-directional uav images

Abstract

With the increasing use of 3D spatial information, updated 3D spatial information is required and unmanned aerial vehicles (UAVs) are being used to accomplish this task. In general, 3D spatial information is constructed by generating meshes through point clouds extracted from many images. However, generating a mesh that is comparable to a real building is difficult because UAV images use relatively cheap sensors compared to aerial images. In this study, Nadir images with various overlapping degrees were obtained and point clouds generated to compare the reproducibility of actual buildings. This led to the confirmation that the use of Nadir images alone does not fully complement the occlusion even when the images are taken with high overlapping degrees. To complement such limitations of unidirectional images, oblique images were used in addition to create point clouds, and the occlusion was found to have been complemented. Furthermore, through its conversion, high quality 3D spatial information data could be obtained.