Researchers have developed a GPS system that analyzes satellite images when planning a route to more accurately determine the type of road and the number of lanes..
Usually GPS maps are created by large companies such as Yandex. and Google, sending special vehicles with cameras to capture infrastructure details in different areas. However, this method of collecting information is expensive and time-consuming, so some parts of the world are simply ignored, and the data obtained is basic. In addition, the system does not always correctly determine the number of lanes and where they lead, so sometimes the route can change dramatically on the way..
One of the options for solving these problems is the use of machine learning algorithms for recognizing roads on satellite images of the terrain. Although this method is cheaper, provides more information and uses regularly updated images, the roads are often hidden by trees and buildings, which complicates the work of AI..
Recently, researchers at the Massachusetts Institute of Technology and Qatar Research Institute jointly developed a combination of neural architectures to automatically predict road types and the number of lanes behind obstacles. During tests, the system correctly identified road types in 93% of cases, and the number of lanes with an accuracy of 77%.
In the future, the team plans to teach the system to recognize parking spaces, bike paths and provide information about current road conditions.
The efficiency of such systems is still important for unmanned vehicles, which goes from test tracks on public roads.
text: Ilya Bauer, photo: today