Google Maps is a triumph of artificial intelligence in action, with the ability to guide us from one place to another using some impressive machine learning technology. But while the routing part of Google Maps doesn’t need too many humans in the mix, manually tracing the roads on the aerial images to make them machine usable is incredibly time-consuming and mundane. As a result, even with thousands of hours spent on this task, Google employees still haven’t managed to map the majority of the 20 million-plus miles of roadways that stretch around the world.
Fortunately, researchers from the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory (CSAIL) may have come up with a solution. They developed an automated method to build roadmaps which is 45 percent more accurate than existing methods. Called RoadTracer, the work uses neural networks to intelligently map roads on images. The system could be especially well-suited to map parts of the world where maps are frequently out of date, like remote and rural areas in the developing world.
“We trained the neural net using aerial images of 25 cities across six countries in North America and Europe,” Favyen Bastani, a graduate student at MIT CSAIL, told Digital Trends. “Specifically, for each city we assembled a corpus of high-resolution satellite imagery from Google Earth and ground truth road network graphs from OpenStreetMap, covering a region of roughly 10 square miles around the city center.”
RoadTracer works by starting with a known location on a road network and then examining the surrounding area to work out what is most likely to be the next part of the road. Once this point has been added, the process is repeated again and again until the entire road network has been added.
Going forward, the team hopes to move beyond relying on principally aerial images for mapping. “For example, they don’t give you information about roads with overpasses, since you obviously cannot see them from above,” Bastani said. “One of our other projects is to train systems on GPS data, and then to eventually be able to merge these approaches into a single mapping system.”
A paper describing the work will be presented in June at the Conference on Computer Vision and Pattern Recognition (CVPR) in Salt Lake City.