Lidar Verisinden Bina Çatı Düzlemlerinin Otomatik Çıkarımı Ve Modellemesi
xmlui.mirage2.itemSummaryView.MetaDataShow full item record
Being as main factor on 3D city models; building modelling is among the most common field of applications of LiDAR (Light Detection And Ranging) point cloud data. In this study, automatic extraction and modeling of building roof planes from the data of 3D airborne LiDAR point cloud dataset of three pilot areas selected from the city center of Bergama / İzmir province is aimed. First, ground filtering process was carried out. Building class was extracted through the classification of the remaining LiDAR points after bare ground points -obtained from ground filtering process- removed from raw data. Following this step, Region Growing Segmentation algorithm was applied on the extracted. Building class and the point cloud of each building was detected separately. Next, the planar surfaces of the building roofs were automatically extracted by applying the 3D RANSAC (3D RANdom SAmple Consensus) algorithm to point cloud of each detected building. After extracting the planar surfaces of the building roofs, the noise points on each building roof plane were identified using the DBSCAN (Density Based Spatial Clustering of Applications with Noise) algorithm and removed from the roof plane points. After removing the noise points, the boundary line was extracted from the points of the building roof plane. As the last step, building roof plane border lines were simplified by using Douglas-Peucker algorithm. When the obtained roof plane models were analyzed it was observed that the best results belong to test field #3. The main reason for having better results in the third test field is that the DBSCAN algorithm is more successful in detecting noise on fewer buildings using the same parameter values. Also, the RANSAC algorithm was more successful in this test field compared to other fields. When the results of test fields #1 and #2 were compared, it was noted that the results of test field #1 were better than the results of test field #2. The main reason for this is that due to lower point density in test field #1 the automatic detection of noise points was performed more successfully.