mind_the_gap: An algorithm for automated gap detection in building footprints data

Keywords: Gap detection, building footprints
Abstract Type: Paper Abstract

Abstract

Building footprints derived from satellite imagery provide a highly useful data source for a wide variety of research topics, such as Oak Ridge National Laboratory’s LandScan program. With recent datasets released by Microsoft and Google, researchers have free access to continent and global scale buildings data. However, these datasets are not complete, resulting in large areas of complete omission and large gaps in their building footprint data. Imagery is typically distributed in square tiles, which if imagery is not suitable for the building footprint extraction, this results in data gaps with a rectilinear pattern. Prior to incorporating these datasets into the LandScan models, these gaps are identified and mapped, but manual gap identification is slow and vulnerable to human error. This is especially a problem in sparsely populated areas where gaps are difficult to visually distinguish. mind_the_gap provides an automated method to map areas of no data in buildings datasets, with a preference for the square or rectangular shaped gaps caused by poor quality image tiles, rather than “true” data gaps such as lakes and forests which typically have a very jagged, irregular shape. mind_the_gap relies solely on building footprints themselves, analyzing their spatial distribution to identify gaps. This algorithm performs well, with near 100% recall when properly tuned and has proved to be an exceptionally useful tool for integrating building footprints data in large scale population modeling, with potential to be used in a wide variety of other applications.