Aerial-terrestrial data fusion for fine-grained detection of urban clues

Abstract

Those who work in the design, development, and management of cities are often limited by the scarcity of data. Particularly in the Global South, urban databases may be insufficient, out of date, or simply not available. However, digital technology is making it possible to fill gaps and build substantial datasets using “urban clues,” or attributes, gathered in high-resolution imagery by sky- and street-based cameras. Aided by machine learning, it is possible to detect specific building characteristics (purpose, condition, size, material, and construction)—yielding an array of geolocated details about the built environment. The resulting composite view can be made available, as we have done, in an open-source portal for use in urban management. The insights gained in this way may help address common urban management challenges, such as locating homes vulnerable to hazards such as flooding or earthquakes, identifying urban sprawl and informal housing, prioritizing infrastructure investments, and guiding public program support. This approach has been applied in Colombia, Guatemala, Indonesia, Mexico, Paraguay, Peru, St Lucia, and St Maarten.