Semantic segmentation of transportation infrastructure.
Our project relies on the GeoNRW dataset, a collection of urban aerial imagery from North Rhine-Westphalia, Germany. It consists of orthorectified aerial imagery, LiDAR-derived digital elevation models, and 10-class segmentation maps derived from German open data in combination with OpenStreetMap data. It contains 7,783 images of each type, and four of its 10 classes represent some type of transportation infrastructure: railway, highway, airport/shipyard, and roads.
For minimizing computational time and to serve as a demonstration, we will only use data from the German city of Cologne (Köln), the largest city in NRW.
The GeoNRW dataset draws from open data used under a German government data license (Data licence Germany – attribution – Version 2.0).