Discover New Roads with Bing Maps

Uncover New Roads with Bing Maps

Posted on

The Microsoft Maps AI Workforce has detected 47.8M km of all roads and 1.16M km of lacking roads from Open Road Maps (OSM). These new roads have been detected utilizing Bing Maps imagery collected between 2020 and 2022 together with sources from each Maxar and Airbus. The whole set of roads is now obtainable to Bing Maps Customers but additionally shared on Github with the open information neighborhood and is freely obtainable for obtain and use underneath the Open Information Commons Open Database License (ODbL).

How did they do it?

Utilizing a Neural Community Structure and dataset in fact! The Microsoft Maps AI workforce’s community was primarily based on UNet and ResNet and the next papers [U-Net] (, [Res U-Net] (, [Res U-Net] ( The mannequin was skilled utilizing the Keras toolkit with Bing Maps 512×512 photographs, it’s absolutely convolutional, which permits photographs of any measurement (which are divisible by 64) to be processed by the mannequin (constrained to 1088×1088 with 100 cm/pixel decision by GPU (graphics processing unit) reminiscence on this occasion). The dataset consisted of 20k labeled satellite tv for pc photographs protecting various areas worldwide. To attain a superb set illustration, the set was additionally enriched with samples from varied areas that included mountains, glaciers, forests, deserts, seashores, coasts, and so on.

As soon as the mannequin was skilled, highway extraction was performed in 4 phases

  1. Semantic Segmentation – Recognizing highway pixels on the aerial picture utilizing Convolutional Neural Community (CNN).
  2. ​Geometry Technology – A collection of algorithms and processes remodeling output of semantic segmentation into roads in geometry format.

    • Picture postprocessing
    • Thinning
    • Connectivity enchancment
    • Graph building;
    • Finalizing highway shapes and community high quality
    • Stitching roads between neighboring photographs (the place wanted)

  3. Conflation & Reducing – Excluding roads and components of roads that exist already within the highway community (OSM).
  4. ​Classification – A classifier to filter out low-confidence roads and predict a highway kind.

How do we all know whether it is good?

The Microsoft Maps AI Workforce measured intermediate stage metrics to trace efficiency of the fashions. This targeted on Pixel metric measures for the efficiency of the Convolutional Neural Community and APLS metric (Common Path Size Similarity) to measure total highway connectivity after the highway geometry technology stage.




Pixel 85.24% 82.81%
APL 87.53% 79.33%

The “Lacking” OSM Information went by means of a last classifier to make sure that the precision is not less than 95%. After classifiers filtered out probably dangerous roads, the precision was remeasured and made certain that it’s 95% earlier than releasing outcomes. 

What does this imply for the long run?

The classic of the roads relies on the classic of the underlying imagery. As Bing Maps Imagery is a composite of a number of sources it’s tough to know the precise dates for particular person items of information however the Microsoft Maps AI Workforce is now linked to the direct imagery replace pipeline for Bing Maps so will proceed to refine and replace this dataset as new imagery is acquired.

How can this information be used?

This set of highway community information is beneficial for a spread of various functions. Having an correct map of rural and concrete roads is a mandatory situation for efficient long-term planning and might save valuable time and assets that will in any other case have been used for surveying entry choices to distant areas.

Probably the most in style functions of our Microsoft Maps AI information, each these new roads and constructing footprints, can been seen in OpenStreetMap. The Humanitarian OpenStreetMap Workforce is successfully crowdsourcing the identification of probably weak areas with the usage of highway community information. As soon as recognized areas are mapped and verified, humanitarian organizations can act shortly in instances of want. 

What format is the info saved in?

Microsoft’s constructing footprint and new roads information is definitely accessible within the GeoJson format, which is usually used to encode a spread of various geographic information. This file format has been chosen on account of its compactness relative to different XML codecs, in addition to its straightforward readability. For instance, here’s a GeoJson file with New York as a reference level.

    "kind": "FeatureCollection",
    "options": [
          "type": "Feature",
              "type": "Point",
              "coordinates": [-74.006393, 40.714172]
              "identify": "New York",
              "description": "New York"

The place can I discover out extra about Bing Maps?

Simply head to to get began and create a Bing Maps API key to find additional functions of our location information and get began with a developer-friendly mapping expertise immediately!

Supply hyperlink

Leave a Reply

Your email address will not be published. Required fields are marked *