On Voxel-Based Sampling of Planar Surface Areas from Photogrammetric UAV Data

Abstract. Photogrammetric methods, when used with UAV-based image acquisition, allow the creation of dense 3D point clouds that can be utilized in numerous applications across various fields. Planar surface areas can be extracted from 3D point clouds and utilized for various purposes, such as building reconstruction, augmented reality, general scene understanding, robot navigation, finding corresponding observations needed for data registration, or in SLAM, to name a few. The aim of this paper was to examine the behavior of voxel-based detection and sampling of planar surface areas from UAV-based photogrammetric 3D point clouds with both non-overlapping and overlapping voxels. As a result, significantly more samples from planar surfaces were found if overlapping voxels were applied. In addition, it was found that the threshold value for detecting planarity needs to be adjusted when the voxel size is changed. The trend was that the threshold needed to be increased when the size of voxels decreased.