RoadTagger AI enhances path recognition accuracy for car navigation

Although global satellite navigation (GPS) systems have been widely used in the civilian market, it is difficult to avoid the embarrassment of turning the wrong junction in the course of daily driving. The good news is that a team from the Massachusetts Institute of Technology (MIT) and the Qatar Computer Research Institute came up with a way to use satellite imagery to enhance existing map data. Nothing attracts us more than the use of artificial intelligence (AI) to calculate the layout of roads obscured by data and buildings.

RoadTagger AI enhances path recognition accuracy for car navigation

(From: MIT, via New Atlas)

The technology, known as RoadTagg, aims to use machine learning on satellite images.

The system is able to pinpoint some additional details on the road(such as how many lanes there are) to provide early warning information such as forks or lane merges.

In addition, RoadTagger can be used to make reasonable estimates of non-motorized roads and parking spaces, making it particularly useful where map data is lacking (additional details are added to the map relatively quickly and at a low cost).

“Large companies tend to provide the latest digital maps in key areas, and small places tend to be overlooked,” says Sam Madden of MIT.

Based on this, the research team decided to focus on the automated generation of high-quality digital maps so that they could be used in any country or region.

During testing in 20 U.S. cities, the RoadTagger system can count lanes with at least 77% accuracy, even if the road’s view is completely or partially obscured.

RoadTagger is 93% accurate in road type identification (residential or highway).

In addition, machine learning systems can identify dirt roads or sheep intestine paths, as well as road features that are obscured by environments such as overpasses.

It is reported that the program AI that supports The RoadTagger can split the road into multiple tiles and invoke information about the surrounding tiles when the view is blocked to help determine the layout of the road.

As an ‘end-to-end’ model, it is able to generate output in raw data without human intervention.

RoadTagger’s application prospects are fairly bright, given that satellite imagery is often updated more timely and fixed frequency than map data.