A Path Guided Method for Quickly Labeling Roads in Satellite Images

Determining roads from satellite images has gained more research interest after the recent progress on data-heavy machine learning methods which are also accelerated by increased amounts of accessible data. An important challenge learning-based approaches is obtaining labeled data to train systems. In this study, we propose a method for quickly labeling over any desired location. Our exploits 2D path information obtained OpenStreetMap, an online community-contributed source geolocated information. environment, roughly described as line segments without their exact shapes and sizes. Using rough information, simple interactive user interface where users easily label road boundaries presented images. our approach, it possible rapidly regions with different characteristics. Such approach allows training separate systems parts world would be advantageous single system identify all kinds roads.