Building detection from 3D point cloud

Building detection from data obtained by remote sensing technologies is one of the most important research topics of our time. It is also important that the building detection process, which is needed in many areas, from population movements to city development, from illegal building observation to casting inference, is accurate and automatic. In this research, the building detection and footprint extraction is made in 4 different study sites by using point clouds data obtained from Light Detection and Ranging (LiDAR) system and Structure from Motion (SfM) with Unmanned Aerial Vehicle (UAV) based aerial images. The basic approach of the study has been adopted as the textures of buildings are different from other objects and this fact can be used in building detection. In the literature survey, it has been found that there are many studies using different data sets and methods in this subject. In this thesis study, the dissimilarity texture parameter which is used for different purposes was used in building determination and contribution to the work in this area was presented. In all these studies, it can be said that common problem plants are mixed with the buildings. While this problem has been solved by using additional data such as vegetation index or classification in other studies, a Morphological Erode operator was used to filter the vegetation with a high rate of success in this study. Applications in different regions with different data sources and the comparison of the results of the Vaihingen data set obtained from ISPRS with the results of other researchers are consistent. As a result, the outcomes of the study obtained by the proposed method, applied in different datasets and test sites with different topography and building properties show that without using any auxiliary data sets, the dissimilarity texture parameter can be used in the building detection studies.