Salzburg / Austria in Pixels

Cloudflight generates elevation models from aerial images, by squeezing trillions of pixels through the point cloud data pipeline!

Castle Hohensalzburg
Shaded digital surface model
Castle Hohensalzburg
3D Mesh, perspective view, b/w
Castle Hohensalzburg
Aerial image, b/w
Castle Hohensalzburg
3D Point Cloud, perspective view

Foreword… let’s go back some years

As you may probably be aware, Cloudflight has its roots – remote sensing-wise – in the observation of Aerosol content of the Earth’s atmosphere. In order to run these highly-specialized and complex algorithms on satellite data from the whole globe you need one thing – desperately: enough computational power.

So already in 2016 we started to build our own private cloud computing infrastructure. If you have a lot of data to process and got the skills you can be much more cost-efficient doing this on your own, rather than renting the whole thing at one of the big players.

The Cloudflight Space Cluster at night is also used for industrial AI applications and telecommunications.
Compute nodes with a total of 1728 vCPUs, 9216 GB RAM; Storage nodes with 3350 TB of disk space!

Fast forward… today

We were thinking: what works so well for the satellite data processing… should also work for earth observation data from other sensors.

In October 2018 we had the opportunity to put this hypothesis to test, when we were tasked by the Austrian Federal Agency for Surveying and Metrology (in short and German acronym: the “BEV”) to run an untypically large demonstration project for them. BEV acquires aerial imagery from the whole of Austria in 3-year cycles. They use these high-quality images – among other purposes – to generate so-called Digital Surface Models (DSMs) and True Orthophotos.

A DSM is a model of the Earth’s surface that depicts the topography and all natural as well as man-made structures on it (houses, trees, hedges, cars, etc.). A True Orthophoto – well, think of a photograph that is free of all distortions that usually arise (e.g. due to projection), so that you can actually measure distances, areas, etc. in it.

Hohensalzburg DSM shading
Hohensalzburg DSM shading
Hohensalzburg TrueDOP
Hohensalzburg TrueDOP

Now, BEV wanted us to generate these DSMs and True Orthophotos for them to find out if we can generate a good quality product faster than they can themselves.

Stay tuned for an update how we tackled 5896 images, 340 Megapixels each… so in total 2 terapixel consuming 18.4 TB disk space…

All data courtesy of the Austrian Federal Agency for Surveying and Metrology. Thank you!


Are you interested in point cloud processing?