Using machine learning to build the future of 3D mapping


Earlier this year we made the alpha release of our building height attributes available to our existing OS MasterMap Topography Layer customers. Almost 20 million building heights across Great Britain were released as an early step in our migration to enhanced 3D geometry and we’ve been gathering feedback ever since.

One of the most common themes being fed into our OS Insight programme has been around the shapes of roofs, as this can help our customers make planning decisions, create realistic views and model sunlight and telephone signals around buildings. Our Research team have started to look into this for the future development of our product.

Currently, our building heights define the bottom of the building, and the top and bottom of the roof. Isabel Sargent and David Holland in our Research team have been working on a small project to see whether it’s possible to automatically extract the shape class of each roof and whether buildings that don’t fit simple height data can be identified.

Using a combination of polygon (vector) and digital surface model (DSM) data, they selected a group of buildings in Hull and ran a cluster analysis.

The results showed around 50 clusters of roofs (shown below), such as pyramidal roofs or L-shaped buildings, and also, buildings which had roofs partially obscured by vegetation. The project now needs to carry out further work to establish what roof shape classes can be identified and how accurate this classification is.


The team will be extending their data set to include other towns such as Eastleigh and Bournemouth to further test their methodology through a mixture of unsupervised and supervised approaches. They hope not only to add the roof shape attribute, but to add at least one further building height attribute to the five already provided in our building height attributes, and to be able to flag buildings that can’t be meaningfully described using the simple attributes and require further investigation. This will all provide vital intelligence in the work on the future of our height products.

Continuing our involvement in 3D, last week we hosted the EuroSDR/ISPRS workshop on ‘Efficient capturing of 3D objects at a national level: with a focus on buildings and infrastructure’. The workshop showed that there is much still to be decided about 3D geographic data such as who will use it, how they will use it and how the data are captured and kept up to date.

In the meantime, the latest release of OS MasterMap Topography Layer – Building Height Attributes is now available. Visit our website to find out more.

You may also like

OS’s adventures in AI
OS Maps has a new 3D fly-through feature
Creating 3D data for Countryfile
3D streaming

Leave a Reply

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

Name* :

Email* :