RANSAC algoritması kullanılarak LiDAR verisinden otomatik çatı düzlemi çıkarılması

Automatic image processing and object extraction from airborne data have become an important topic of research in the field of photogrammetry and remote sensing. The aerial laser scanning system, also known as LiDAR, has become the dominant technology for acquiring 3D spatial data from the earth surface with high speed and density. LiDAR's output is an unclassified and unstructured point cloud dataset. Thus, the main process to be performed on this dataset is to classify it into distinct classes. Then, the classified LiDAR data can be used as input to create 3D city models. This data has a number of unique properties that play a fundamental part in their classification process. The main properties include the geometric properties that are obtained through the processes carried out on 3D positions of the points in the cloud. Among these processes is the plane extraction, which is carried out through the most commonly used methods of RANSAC (Random Sample Consensus), Region growing, and Hough Transform. In this study, the RANSAC algorithm was used to extract planes from building rooftops. The aim is to apply RANSAC on LiDAR point cloud data to extract planes from rooftops. The first and most important step in the extraction process of the planes from rooftops is to identify and distinguish buildings from the other features, such as terrain and vegetation. The second step is to apply the RANSAC algorithm on the point cloud data of the individual buildings. Based on the geometric position and the points' distance to the plane, the least squares method is used to cross the best plane through the candidate points that form the plane. The experiments were carried out on the selected study areas located in the city of Bergama, Turkey using LiDAR point cloud data collected by the Reigl airborne scanner. The results show that RANSAC's performance is quite good for buildings which have complex roofs and also it has the ability to extract small planes in high density point clouds. Furthermore, the best extracted planes are properly adjusted to the raw point cloud data sets.