Automatic building extraction and digitalization through photogrammetric image based point cloud

Determining the building boundaries manually from terrestrial method or remote sensing data is time-consuming and demanding. Building boundary extraction from relevant data sources is extremely important as it is a map base. Point clouds produced by matching and laser scanning provide dense and highly accurate three-dimensional (3D) position information. Automatic extraction of buildings from the 3D point cloud is a very difficult problem. Automatic data processing contributes significantly to reducing processing time, cost and operator error in map production. In this study, it is aimed to automatically extract and digitize building details by using point clouds produced from photogrammetric images and LiDAR point clouds for comparison purposes, with the new approach (improved Octree, I-Octree) developed and proposed by automating the working principle of the Octree data organization method. To achieve this goal: (i) generating a large volume of 3D point cloud from images or providing a LiDAR point cloud; (ii) separation of ground and above ground objects from the point cloud by SMRF; (iii) classification of building objects by DBSCAN algorithm; (iv) extracting building details by applying the octagon and improved octree (I-Octree) method to classified objects; and (v) smoothing building edges with ABORE is focused. The proposed approach was applied on photogrammetric point clouds in Elazig and Erzurum test area and on LiDAR point cloud in California test area. Reference maps with scales of 1/1000 and 1/5000 were used as reference data for all test areas. Object-based validation was performed with completeness (Cp), accuracy (Cr), quality (Q) and F-score (F-1) metrics. In the test areas, the verification results are above a maximum of 94% for each metric in photogrammetric point clouds, while this value is above 99% in the LiDAR point cloud test area. As a result of the validation, it is seen that the proposed approach can extract the selected buildings in the test areas with high accuracy. As a result, it has been revealed that with the proposed approach, building detail extraction and digitization can be performed automatically from point clouds of different densities.