Fine-grained population estimation

Abstract

We show how to estimate population numbers for arbitrary user-defined regions, down to the level of individual buildings. This is important for various applications like evacuation planning, facility placement, or traffic estimation. However, census data with precise population numbers is typically only available at the level of cities, villages, or districts, if at all. Previous approaches either rely on available census data for already small areas or on sophisticated input data like high resolution aerial images. Our framework uses only freely available data, in particular, OpenStreetMap data. In the OpenStreetMap project, crowd-sourced data is collected about street networks, buildings, places of interest as well as all kind of regions and natural structures world-wide. We use this data to learn three classifiers that are relevant for the population distribution inside an area: residential vs. industrial vs. commercial landuse, inhabited vs. uninhabited buildings, and single-family vs. multi-family houses. Once learned, we can use these classifiers for population estimation even in areas without any census data at all. Our experiments show good average accuracy (measured as the deviation from actual census data) for rural areas (25%), metropolitan areas (10%), and cities in countries other than that containing the training data (12%).