GESTALT: Geospatially Enhanced Search with Terrain Augmented Location Targeting

Geographic information systems (GIS) provide users with a means to efficiently search over spatial data given certain key pieces of information, like the coordinates or exact name of a location of interest. Current GIS capabilities do not enable users to search for locations using imperfect or incomplete information easily. In these cases, GIS tools help narrow down a region of interest, but users must conduct a manual last-mile search to find the exact location of interest within that region. This typically involves the user visually inspecting many remote sensing or street-view images to identify distinct landmarks or terrain features that match the partial information provided. This step of the search process is a bottleneck. Taking inspiration from the way humans recall and search for information, we present the Geospatially Enhanced Search with Terrain Augmented Location Targeting (GESTALT), an end-to-end pipeline for extracting geospatial data, transforming it into coherent spatial relations, storing those relations, and searching over them. We contribute a new Swan Valley Wineries dataset and a proof of concept architecture that includes multiple methods for querying spatial configurations of objects, handling uncertainty in the information known about a location or object, and accounting for the fuzzy boundaries between locations.