Researchers from Facebook and MIT Labs have proposed a new methodology that uses machine learning and satellite imagery to generate street addresses in areas of the world where individual buildings don’t have a unique address. The methodology divides the street addressing into two processes. The first process is segmentation. During segmentation, road pixels are identified using deep learning from 0.5 meter resolution satellite images. The second part of segmentation involves developing the road network from these identified pixels. Next, the road network is divided into regions. The second process is called labeling. During this process, the regions, road segments, and place markers are named and block letters are assigned to each unit. The regions are divided into quadrants (N, S, E, W) with the city centered defined as the densest area. The streets are numbered and lettered based on their proximity and orientation from the centre of the city. In comparing the results of the model with manually labeled GIS data, the researchers found that the model was able to learn on average 80% of the roads per city.
There have been other efforts to develop an addressing system for regions of the world that don’t have street addressing. What3Words is one such endeavor which assigns a random combination of three words on a grid system of the world that has been divided into 3 meter by 3 meter cells. What3Words has been adopted by the postal system in Mongolia to help with mail delivery.
Demir, I., & Raskar, R. (2018). Addressing the Invisible: Street Address Generation for Developing Countries with Deep Learning. arXiv preprint arXiv:1811.07769.
Four billion people lack an address. Machine learning could change that. MIT Review