Automatically and Accurately Conflating Raster Maps with Orthoimagery

Recent growth of geospatial information online has made it possible to access various maps and orthoimagery. Conflating these maps and imagery can create images that combine the visual appeal of imagery with the attribution information from maps. The existing systems require human intervention to conflate maps with imagery. We present a novel approach that utilizes vector datasets as “glue” to automatically conflate street maps with imagery. First, our approach extracts road intersections from imagery and maps as control points. Then, it aligns the two point sets by computing the matched point pattern. Finally, it aligns maps with imagery based on the matched pattern. The experiments show that our approach can conflate various maps with imagery, such that in our experiments on TIGER-maps covering part of St. Louis county, MO, 85.2% of the conflated map roads are within 10.8 m from the actual roads compared to 51.7% for the original and georeferenced TIGER-map roads.