Building footprint extraction using orthophotos based on Artificial Neural Network and fusion of dense point cloud with Digital Topographic Map — Istanbul, Turkey

Buildings are important structures that form the fabric of the city, especially in metropolitan cities. The science of remote sensing has several advantages in the extraction of buildings, which included the urban model. Light Detection and Ranging (LiDAR) is an important data source that has become increasingly common over the past few years, and is frequently used in urban planning. On the other hand, Artificial Neural Network (ANN) technology is a model that consists of many neurons which is inspired by the human brain. However, it is not always possible to make high-accuracy level building extraction in one step with this data. In this study, firstly, for enhancing more in-depth analysis, building footprint extraction was carried out by combination of high-accuracy Regularize Digital Topographic Map (RDTM) with LiDAR data in urban area. Secondly, to reveal the efficiency of orthophoto in building detection using ANN method. The derived LiDAR model and the combined RDTM models are the most efficient in classifying buildings. In this model, Correctness (CR) is 98%, Completeness (CP) is 97%, Quality (QL) is 96% and F-score is 97%. With respect to these research findings, the LiDAR-RDTM fusion model has proven to be a very effective method in complex urban areas. The ANN method which used only orthophoto has less performance compared to LiDAR/RDTM combination, and Overall Accuracy (OA) 90.15%, and kappa coefficient is 0.83.