Building extraction from data obtained by unmanned aerial vehicle

Nowadays, parallel to technology, data collection technologies have also improved. Remote sensing platforms such as satellite, lidar and unmanned aerial vehicle (UAV) and data collection tools have been used in various studies. Very dense data can be collected quickly with such data collection equipment. It is necessary to determine the meaningful ones from the dense data obtained. Many details need to be followed especially for changes in rapidly developing residential areas. For this purpose, the desired details should be extracted from dense data sets. In this study, application was carried out on the extraction of the buildings by using point cloud and orthophoto images produced from the images obtained by unmanned aerial vehicle (UAV). Within the scope of the study, it was aimed to determine the buildings in the Afyon Kocatepe Campus in Afyon. The rule sets used in the study were then applied in the test data. Selected datasets were evaluated in photogrammetric software and medium and high density point clouds and orthophotos were produced. Point clouds of different density produced are classified in Pix4D, Quick Terrain and VRMesh software. By using object-based classification approach to orthophoto images, building detection was made in eCognition software. In the classification of object-oriented applications made in the eCognition software, objects are divided into two classes: building and non-building. An accuracy analysis was conducted to investigate the accuracy of the buildings identified as a result of the study. As a result of the study, accuracy analyzes of orthophotos and point clouds were compared. The analysis of different density data produced from the Afyon dataset was analyzed. The highest quality value of medium density data was found in Pix4D software with 78.69%, and the highest quality value of high density data was found in VRMesh software with 84.84%. In the Swiss dataset, the medium and high density dataset was determined as 80.01% and 85.32% in VRMesh software. In orthophoto images; the quality of the Swiss dataset was better than the Afyon dataset with 92.64%. As a result of the study, it was observed that the classification made from orthophoto images gave more accurate results than the point cloud data.