html templates

Machine learning for map makers

Hidden insights revealed in pictures taken satellite pictures

Images taken from Earth orbit can show places all round our planet's surface. We can spot roads, junctions, landmarks, and even navigate via photos containing street names. Sometimes it feels as though there are no places left unexplored. 

But in much of the world, many roads are still not labelled on maps. Small routes may be missing, even when the main road is identified. People who live in such countries, and organisations visiting them such as the Red Cross, want a way of filling in the gaps. 

Nobody can go through every satellite image. So a Computer Science MSc student at University College London worked with IBM technology and an IBM Senior Inventor to create a "Missing Map" project that can  detect the roads and building that appear in satellite images. 

The Python application was developed in IBM Watson Studio. It uses IBM Watson Visual Recognition for a ready-to-go model that does not require a detailed knowledge of machine learning to achieve an appropriate level of precision when it connects to the IBM Cloud. 

Subsequently, a web service will let people access the application more easily, and additional training for the machine learning will extend the app to different land areas such as desert, grassland, and savanna. 

Come and see how the app can highlight roads and buildings from raw aerial images, even though they are not yet labelled on familiar services such as Google Maps. 


IBM UK Lab Campus, Hursley,
Winchester, Hampshire,
SO21 2JN
Email us