Using deep learning and Open Street Maps to find features in aerial images

Abstract

[en] A great amount of the interesting information captured by aerial imagery is still not being used given how labour intensive the processing and annotation of these images is. Despite this, improvements in technology and advancements in the computer vision field have made available tools and techniques that can help make this process semi-automatized. In this project we focus on the use case of extracting roads from aerial imagery. For this purpose, we will study and compare models based on image segmentation using deep learning and RoadTracer, a revolutionary model proposed recently.