camera model export

113 views
Skip to first unread message

Curtis, Aaron (347C-Affiliate)

unread,
Jan 31, 2018, 2:56:07 PM1/31/18
to ames-stereo-pi...@googlegroups.com

Hello, I'm experimenting with combing ASP with commercial multiview photogrammetry tools (Agisoft Photoscan, Reality Capture, Pix4D). I'm working with Lunar Reconnaissance Orbiter NAC imagery and would like to be able to export camera locations / models to a format that can imported by those programs. Has anyone done this successfully? I see that cam2rpc exports camera models to an xml format, but it is different from Photoscan's camera xml format. Ideally, I would like to export cameras from ASP to either:

 - Bundler .out format

 - BINGO .dat format

 - Realviz RZML

 - VisionMap .txt


Because these are the formats that Photoscan can import.


Thank you,

Aaron Curtis


NASA postdoc at JPL

347C Extreme Environment Robotics

Alexandrov, Oleg (ARC-TI)[SGT, INC]

unread,
Jan 31, 2018, 3:55:48 PM1/31/18
to Curtis, Aaron (JPL-347C)[Affiliate], ames-stereo-pi...@googlegroups.com
Dear Aaron,

I am glad to hear you are doing this kind of experiments, and if later you develop any insights you would like to share, we would be curious to know about that. 

Regretfully we do not provide any export facility to any of these formats. And that may not even be possible. You see, LRO NAC is not a simple frame or pinhole camera, like a square matrix of pixels used in commercial photography. LRO NAC is a linescan device, meaning that the sensors are an array, and as the satellite moves, another array is created under it, and so on. The imaging model is a lot more complex. For example, the distance between pixels in x and in y is not the same, in fact the two can differ by a factor of 4 or so and can vary across the image. Lens distortion is also not working the same way. 

The RPC XML model we export with cam2rpc captures that complexity, and that is the standard used in satellite imaging, whether with American or European (and even other nation) providers. 

That being said, it may be possible to approximate LRO NAC's images with some kind of fake pinhole model, but I don't know how easy that would be or how accurate. 

In short, you are trying to use linescan satellite imagery with packages geared towards aerial photography done with frame cameras, and the two don't mix too well. 

I hope you can find a solution.

Oleg






From: ames-stereo-pi...@googlegroups.com [ames-stereo-pi...@googlegroups.com] on behalf of Curtis, Aaron (347C-Affiliate) [aaron....@jpl.nasa.gov]
Sent: Wednesday, January 31, 2018 11:56 AM
To: ames-stereo-pi...@googlegroups.com
Subject: camera model export

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/0962eb9c55294e7aba63ccb73ef3dde4%40jpl.nasa.gov.
For more options, visit https://groups.google.com/d/optout.

Curtis, Aaron (347C-Affiliate)

unread,
Jan 31, 2018, 5:37:52 PM1/31/18
to ames-stereo-pi...@googlegroups.com

Thanks for the info, Oleg! Yes… there are a lot of people on the Photoscan forums begging for PS to add support for pushbroom sensors.

 

I’m interested in doing stereo on fairly small areas of NAC images ( < 10km x 10km ) – do you think at a small scale there is a reasonable way to pretend I have a pinhole or frame camera?

 

Even if not, it would still be very helpful for me to have a rough idea of the instrument location in space during the photo. I guess that is something I could extract from SPICE data if I knew how to, right?

Alexandrov, Oleg (ARC-TI)[SGT, INC]

unread,
Jan 31, 2018, 5:52:43 PM1/31/18
to Curtis, Aaron (JPL-347C)[Affiliate], ames-stereo-pi...@googlegroups.com
Aaron,

I hope that the Photoscan people will add support for pushbroom. I would be a bit skeptical though, as the pinhole camera assumption is at the very core of what makes Photoscan's Structure-from-Motion work. They would have to rewrite their entire code to add support. And they may not have a good reason, as pushbroom sensors are are used for satellites only, the images are immensely huge, not fitting in memory, which will also break their tool. Also, Photoscan's functionality may not be needed for such images, since satellite vendors whose telescopes cost hundreds of millions of dollars do a much better job than your average hobbyist with a DSLR in his hand at providing camera meta info so that structure from motion may not be necessary. In short, good luck! 

For NAC images, 10 km by 10 km may not be small. I think their images are 2.5 km pixels wide or maybe 5km. I'd say "small" can be something like 2 km by 2 km or even 1 km by 1 km. And again, you'd have to keep in mind that pixel width in x and y are different. There is also jitter, as the line sensor wiggles a bit, and the frame camera won't account for it. 

Yes, the SPICE data has the info you want. We never deal with SPICE directly, and I don't know how to query it. There is a chance the tool campt may help (https://isis.astrogeology.usgs.gov/Application/presentation/Tabbed/campt/campt.html)  but I never used it. 

Oleg


Sent: Wednesday, January 31, 2018 2:37 PM
To: ames-stereo-pi...@googlegroups.com
Reply all
Reply to author
Forward
0 new messages