top of page

AI Case Study

Facebook create new location address schemes for developing areas decreasing travel times by 22% in tests

Facebook researchers develop a model of predicting roadways and other human infrastructure from satellite images in order to create an accurate addressing system, particularly for areas where no consistent system exists. The goal is to aid developing countries and improve direction accuracy to help disaster relief. The automatic addressing system covers more than 80% of populated areas.


Public And Social Sector


Project Overview

The researchers "designed and implemented a generative addressing system to bridge the gap between grid-based digital addressing schemes and traditional street addresses. We (i) design a physical addressing scheme, which is linear, hierarchical, flexible, intuitive, perceptible, robust, (ii) propose a segmentation method to obtain road segments and regions from satellite imagery, using deep learning and graph-partitioning, (iii) implement a labeling method to name urban elements
based on current addressing schemes and distance fields, and (iv) develop ready-to-deploy prototype applications supporting forward and inverse geoqueries."

Reported Results

"We compare the road predictions with ground truth for the extracted roads of an unmapped suburban area. Our SegNet model and post-processing approach were able to learn 90.51% of the roads. This success ratio was close to 80% on average per city. We accomplished automatically addressing more than 80% of the populated areas, which significantly improved map coverage."


"The first step of our approach creates binary road prediction images from three channel satellite images of 0.5 m resolution and of size 19 K * 19 K. Both training and testing are done with patches of 192 × 192. The training set includes 4–16 tiles per country, and the test set includes all the rest of the tiles, manually spatially distributed to sample all areas, keeping the ratios mostly at 70% to 30%. We use a modified version of SegNet [6], which consists of the first 13 convolutional layers of the VGG16 network for the encoder, having a corresponding decoder layer for each encoder. We modify the last soft-max layer to change the multi-class structure to have binary classes for road detection, by substituting it with a convolutional layer."



Field Services


"Street addresses enhance precise physical presence and effectively increase the connectivity all around
the world. Currently 75% of the roads in the world are not mapped, and this number is increasing in developing countries. United Nations claims to have 4 billion invisible people in the world due
to the addressing problem in developing countries. This problem is even more critical in disaster zones, areas with limited resources, and geographically challenging locations. As the remote sensing
technology has been significantly improving over the past decade, the organic growth of urban and suburban areas outruns the deployment of addressing schemes."



Three channel satellite images

bottom of page