Robust road topology extraction from aerial imagery

Abstract

Creating and updating road maps is currently an expensive and often manual process, and thus maps today are outdated or have poor coverage in large regions of the world. Automatically inferring the road network graph from aerial imagery provides a promising avenue to reducing the cost of maintaining road maps, but existing inference methods have poor precision. This thesis develops a novel iterative graph construction process for extracting graph structures from images, and applies this process to automatic road topology inference to significantly reduce error rates.