Researchers blame Google Maps for misleading Uber drivers and have created their own AI model to count roads

An AI model developed at MIT and Qatar Computing Research Institute that uses only satellite imagery to automatically tag road features in digital maps could improve GPS navigation, especially in countries with limited map dataGoogle Maps/MIT News
  • Researchers from MIT and the Qatar Computing Research Institute (QCRI) have a new road-mapping model that can automatically count lanes and detect road types.
  • The researchers claim that Google's current methods to make maps detailed are expensive and often ignore entire parts of the world.
  • RoadTagger uses artificial intelligence (AI) and machine learning (ML) on satellite images to make the process cheaper.
  • They hope that it can help improve maps in countries like Qatar, who is set to host the World Cup 2022 but Uber drivers frequently get lost.
It's not uncommon for dependable mobile navigation apps, that normally work perfectly well in the city, to completely lead you off-the-road if you're exploring unfamiliar territory. However, researchers have a new way to add lane counts and road types to mobile maps to make routes more accurate.

"Most updated digital maps are from places that big companies care the most about. If you're in places they don't care about much, you're at a disadvantage with respect to the quality of the map," said co-author of the study, Sam Madden.

The researchers point out that tech giants like Google do try and be as detailed as possible — but the process is expensive. In some cases, entire parts of the world are ignored. In India, it's a frequent problem when driving through rural areas.

However, the new model — ' RoadTagger' — is cheap, automatic and can be done by simply using satellite imagery. By using machine learning (ML) models on satellite images, which are easier to obtain and can update fairly regularly.

RoadTagger was presented at the Association for the Advancement of Artificial Intelligence (AI) conference by MIT and the Qatar Computing Research Institute (QCRI). It uses a combination of neural network architecture to automatically predict the number of lanes and road types — residential or highway.

Example of RoadTagger detecting the number of lanes and road typesMIT

Navigating FIFA World Cup 2022
Qatar, where QCRI is based, is set to host the FIFA World Cup in 2022. According to Madden, it is "not a priority for the large companies building digital maps.". What makes things more difficult is that in preparation for the tournament, the country is constantly building new roads and improving on old ones.

"While visiting Qatar, we've had experiences where our Uber driver can't figure out how to get where he's going because the map is so off," said Madden.

By integrating lane counts, the GPS system can warn drivers of diverging and merging lanes. Information on parking spots could help drivers plan ahead. Even in cases of disaster relief, tagging roads could quickly update information on road conditions to help improve planning.

"If navigation apps don't have the right information, for things such as lane merging, this could be frustrating or worse," explained Madden.

RoadTagger has been tested using maps of 20 US cities. Researchers found that the model was able to detect lane numbers with 77% accuracy and road types with 95% accuracy. What makes it special is that it can detect roads, even if there are obstructions in the way, like overarching trees that may make parts of a road disappear from satellite images.

Two representative samples of the micro benchmark, RoadTagger predicts both of them correctlyMIT

How RoadTagger works
The RoadTagger model combines a convolutional neural network (CNN) with a graph neural network (GNN).

CNN first puts the raw satellite images of the target roads. The GNN then divides detected roads in 20-meter segments — or 'tiles'. Each tile is then analysed by CNN and compared to the tiles around it. If there's an obstruction, RoadTagger used combines information from other tiles along the road to figure out what's behind it.

Overview of RoadTagger road attribute inference frameworkMIT

The model is 'end-to-end'. This means it operates without human intervention from start to finish.

Researchers hope that RoadTagger will help others validate and approve modifications to more mainstream maps. In addition to Qatar, Bastani in Thailand is another area of interest where the roads are constantly changing but the map updates are few.

See also:
Google I/O 2019: Google won't be stalking you anymore with Google Maps' incognito mode

Google Maps lets you explore the cosmos like Star Trek — complete with a warp drive 'whoosh'

{{}}
Add Comment()
Comments ()
X
Sort By:
Be the first one to comment.
We have sent you a verification email. This comment will be published once verification is done.