Yüksek çözünürlüklü nokta bulutlarından çoklu çözünürlüklü düzlemsellik tabanlı yaklaşım kullanarak dijital arazi modeli çıkarımı

Digital Elevation Model (DEM) is a mathematical representation of the elevation of the Earth's surface. There are two types of DEM, namely Digital Surface Model (DSM) and Digital Terrain Model (DTM). DSM contains natural (bare-ground, trees, bushes, etc.) and artificial above-ground objects (buildings, vehicles, powerlines, etc.), while DTM covers only the bare earth without anything on it. Above-ground objects need to be removed to extract the DTM, which is a tedious task. This thesis proposes an algorithm that extracts DTM from aerial point clouds using a robust multi-resolution planarity-based divide-and-conquer algorithm. In this approach, the problem is handled in few simple steps rather than trying to solve the problem at once. The approach contains different planarity checks to get rid of nonplanar above-ground objects, segmentation step to find rough ground points, and an interpolation step to obtain the final DTM. In this thesis, ground points are assumed planar, and planar patches are detected as ground candidates. First, approximate planarity values are calculated by using neighboring points. This helps to eliminate most of the above-ground objects such as vehicles, trees, posts, etc. Nevertheless, since the building facades and roofs are also planar, a second planarity check is needed in different resolutions. For this purpose, the grid planarity values are checked. The grids that do not fit a plane within the given threshold are marked as nonplanar. The second planarity check helps to get rid of the building facades and the vertical planes. After removing building facades, getting benefit from the sparsity between ground candidates and the roof points, a region growing segmentation is utilized to segment the remaining ground candidates for rough ground surface calculation. The segments far from the rough ground surface are omitted. By doing so, the roof points can be eliminated. Lastly, the ground points are interpolated to obtain the resulting DTM raster. Although the input point cloud is already classified as ground and non-ground, it has some errors. The input point cloud is used to create a DTM; then, the resulting DTM is manually edited to use it as a ground truth. The accuracy assessment is done on interpolated DTM rasters. Using a manually corrected ground truth, Root Mean Square Error (RMSE) is calculated for two datasets with different characteristics having 1.00 m and 2.20 m spatial resolutions. The results are compared with two existing DTM extraction algorithms and RMSE values are close to these solutions. The RMSE values are 0.25 m and 0.70 m, respectively. Results indicate that an accurate DTM extraction is possible using a combination of only planarity values. Keywords: Digital terrain model, digital elevation model, planarity, covariance features, principal component analysis, point cloud