Read previous article: Salzburg / Austria in Pixels
The BEV (Austrian Federal Agency for Surveying and Metrology) wanted us to generate Digital Surface Models and True Orthophotos for them to find out if we can generate a good quality product faster than they can themselves.
So we had everything in place:
- A powerful compute cluster:
- Compute nodes with a total of 1728 vCPUs, 9216 GB RAM
- CUDA nodes with a total of 82 vCPUs, 336 GB RAM, 40064 GPU cores, 115 GB GPU RAM
- A large storage cluster: 560 vCPUs, 3328 GB RAM, 3528 TB HDD storage
- All set up in our data centre: 10 GBit SFP+ and 40 GBit QSFP+ network connectivity, VIX Port
- 5896 images, 340 Megapixels each… so in total 2 Terapixel consuming 18.4 TB disk space
However, we still needed an important part of the puzzle: the method to actually reconstruct the DSMs and create True Orthophotos from the delivered aerial images. A process that is far from trivial and requires highly-specialized domain knowledge to accomplish.
For that we decided to team up with an already well-established and esteemed group doing exactly that: nFrames.
Their software SURE Aerial delivers precise large scale point clouds, elevation data, True Orthophotos and more. We integrated SURE in our processing cluster environment and off we went! Utilizing our cluster we were able to create the end product for the customer BEV in 4.8 days.
This included the DSMs, True Orthophotos, as well as exports of the 3D point clouds with True Orthophoto colours – which are a nice byproduct in the process – see this example of the city of Salzburg for yourself: https://salzburg-3d.cloudflight.dev/
Or Try the Interactive Demo!(Optimized for Google Chrome. Mobile clients currently not supported)
The performance we observe means a stunning throughput of more than 400 gigapixels per day.
At this rate, we are able to finish the whole of Austria – which is approximately 40 terapixel from 3 full years of image acquisition – in less than 100 days, thus being roughly 10-times faster as the current process.
All data courtesy of the Federal Agency for Surveying and Metrology. Thank you!