Data-driven Urbanism: Image Processing Techniques for Urban Analytics

Geographic databases provided by open and public sources lack a high degree of accuracy. Although these sources were developed by collecting data from surveys, tracing from aerial imagery and freely licensed geodata sources, their reliability is questionable in testing new concepts for urban analytics and developing solutions for City Information Modelling (CIM). This study aims to examine a method using digital image processing to deliver precise information and accurate data for urban analytics. The research applied algorithmic solutions using content-based image segmentation, which accurately segments roof regions of buildings from aerial images. The study utilised an open access dataset annotated using 72 images grouped into 6 larger tiles from a joint project between Humans in the Loop with the Mohammed Bin Rashid Space Centre in Dubai, the UAE. The results show the efficiency of extracting buildings and their detailed features in an urban context. Finally, the study demonstrates the reliability of using the Base UNet model and the ResNet-based UNet, in analyzing urban aerial images.