We are using Context Capture Desktop edition 20.1 V10.20.1.5562 and noticed that there have been some updates to the AT engine and an additional low level preset added.
I was hoping more information could be shared about these settings. What criteria needs to be met to trust the metadata? Is this position and orientation data from EXIF/imported data? Is position only sufficient? Is the metadata only trusted more when the low level preset is used, or does this happen automatically when metadata meets the criteria?
What is the difference between using the low level preset vs setting the AT to use Photo positioning metadata as an adjustment constraint as adjust/adjust within tolerance?
Thanks,
Jason.
Hello Jason,
This modifiction will only have an effect if adjustement constraint on "photo positioning metadata" is selected in the AT settings. the photo positioning metadata can come from either EXIF information or imported from a CSV file. This update pushes back the threshold for which the engine consider (rightfully or not) that the photo positionning metadata is trustable or not.
This preset will only have an effect if adjustement constraint on "photo positioning metadata" is selected in the AT settings. This is basically a more drastic version of the default behaviour expalined in the bullet above. In this case, the engine will alsways consider the position metadata as trustworhty and will try to pull on the other parameters during the bundle adjustement. While this can help with the most difficult cases of curved AT, it can also introduce breaks if the input position is not 100% reliable.
I hope this helps !
Sylvain
Answer Verified By: Jason Hagon
Hi Sylvain, fantastic, thanks for the detailed description. We have run a few tests and noticed a significant improvement.
That is great to hear ! Thanks. Could you describe what type of improvements you have experienced, for which type of projects ? Thanks
Prior to this update we would often run an AT using either using Photo positioning metadata - compute or adjust. More often we would import accurate PPK image positions, but these did not contain orientation metadata. This would result in only compute being available. Often at one end of a AT block we would see images moved from their initial position by around 5metres. This could sometimes be corrected by using GCPs if they were in the same area, or a series of additional ATs using the adjust option. Alternatively, using both accurate position and orientation metadata we would run an initial AT using Photo position metadata - adjust. Often this would result in a low number of images being calibrated. Consequently a second AT would be run using the extend option. As this uses the default parameters during the 2nd AT, this would also sometimes result in AT block deformation.
Using the new AT settings on the same images - a grid pattern flown along a 2 km stretch - saw all images calibrated initially within the expected position difference from metadata < 5cm.
Thanks. Jason.
Is this also applicable to issue which is logged in ST7001479154 regarding Wingtra position import?
Hello Oto,
I have reviewed this SR and I hope you have recieved an answer from the support team by now. This case is not really related to the improvments mentionned above. For your specific dataset, there are 3 things :
- there is no added value to import the position and rotation using the CSV file, because the exact same infomation is stored in the EXIF.
- the guide in the Wingtra Knowledge base gives a wrong indication on the setting to be used for importing the orientation metadata. I have reached out to the Wingtra team to get this corrected.
- The input orientation is off by 6deg on average, which makes the AT to reject most photos if "adjust" is set on position and rotation in the AT settings (if set to "compute" it works well). We are investingating if this is an issue of how we read the rotation information, or if it is simply a bit too inncurate for our current AT engine to work with it properly.
This said, we are still working making progress on the new AT engine, to have additional robustness to this, and some failsafe to get a succesful AT regardless and not be so dependant of how the settings were choosen in the first place.