Thomas Koch

Performance for UAS processing (TBC PM / UASMaster)

Discussion created by Thomas Koch on Apr 15, 2016
Latest reply on May 11, 2016 by Thomas Koch

As this topic pops up regularly a quick note on some facts.

1. Often, users wonder why in UASMaster, processing a data set for the first time takes about 10 times longer than re-processing it a second time.

- Reason is that UASMaster stores "raw" measurements from the SFM (structure for motion) matching part. When just doing the tie point extraction again, without physically deleting the raw data, UASMaster takes the raw measurements into account and just processes image measurements out of that. This can be done at a fraction of the time!

 

2. Some users wonder, why their old projects were running faster than newer ones in UASMaster and/or TBC - even if they have approximately the same number of images and same settings. In most of those cases we figured out that they were e.g. using data acquired with an older UX5 previously (which had a 24MPix) camera instead of e.g. the UX5 HP with a much higher resolution. If you would regard the performance per pixel, the performance would be almost identical. In fact, however, with the UX5HP, the software has to process way more pixels and therefore the computation time gets considerably longer.

 

Of course there are a lot more reasons why processing may take longer or shorter (e.g.: in most cases there is almost no difference of the quality of the georeferences if users are doing the tie point extraction the half-resolution level, only). In this thread I just wanted to highlight two very obvious reasons that users constantly getting trapped in.

Other than that, our engineering is constantly improving strategies and algorithms to keep accuracy getting better and computation speed getting higher.

Thanks

Thomas

Outcomes