bundle_adjust on More than 2 LROC Images with Variable Illumination

[Mayer, David P]

Hi Folks,

I'm trying to jointly bundle adjust ~10 overlapping LROC images with the goal of eventually running them through sfs​, but image-to-image alignment seems to be worse after running bundle_adjust. My basic workflow involves running lronac2isis.py​ on the Left and Right detector images from each observation, and then 

bundle_adjust --num-passes 2 <image_1 image_2 image_3...> -o <ba_prefix>​

The list of 10 images is actually made up of 5 targeted stereopairs, so I process each stereopair with parallel_stereo​ using the bundle adjustment result obtained from all 10 images, followed by point2dem​. I then mapproject​ the images from each stereopair onto the corresponding DTM using the appropriate bundle adjust prefix. When I go to look at the orthorectified images in QGIS, images that come from different stereopairs seem to be offset from one another by ~50 meters.  Hillshades of the underlying DTMs appear to have similar offsets from one another, which suggests that bundle_adjust didn't actually do a very good job of registering images from different stereopairs to each other.

The images from different targeted stereopairs do have different solar azimuths, and I'm wondering if that's throwing off bundle_adjust​, but I don't really know  I did save the control network as a CSV and when inspecting it, I noticed that there are tie points between images belonging to different stereopairs, but I don't see anything resembling tie points measured in 3 or more images. Is there a way to force bundle_adjust​ to establish tie points in 3 or more images? Is this a red herring?

Does anyone have experience trying to register LROC images under similar conditions? Are there other automated approaches I can take to get the images bundle adjusted together, or am I stuck picking tie points manually?

Cheers,
David

-----------

David Mayer

Geographer

Astrogeology Science Center

US Geological Survey

dpm...@usgs.gov

[Alexandrov, Oleg (ARC-TI)[KBR Wyle Services, LLC]]

David,

Bundle adjustment of images with different illumination conditions is tricky. It is possible, but for best results one should sort the images by the solar azimuth, so each image can be matched with neighbors of similar illumination. If you run 

sfs -i dem.tif image*cub --query

 it will print on the screen the azimuth, then they can be sorted in the order given by these azimuths. 

A simple visual check is to pick two images with similar azimuth, run bundle adjustment on both of them, and examine the match file. It is also suggested to use the bundle adjustment option --mapprojected data to correct for perspective before interest point matching happens. That can help a lot. You can also examine the images after mapprojection to see if they look similar enough to the eye. 

I will also suggest you also read the documentation in 

https://stereopipeline.readthedocs.io/en/latest/sfs.html

including section 11.5.

The short summary of things is that LRO NAC has a wealth of images, and the bundle adjustment and sfs tools were tested rather exhaustively, so it will work. But, it will require many images with gently varying illumination conditions and a lot of patience. Let me know how it goes.

Oleg

---

From: 'Mayer, David P' via Ames Stereo Pipeline Support <ames-stereo-pi...@googlegroups.com>
Sent: Monday, April 12, 2021 1:40 PM
To: ames-stereo-pi...@googlegroups.com <ames-stereo-pi...@googlegroups.com>
Subject: [EXTERNAL] bundle_adjust on More than 2 LROC Images with Variable Illumination

 

--
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/BY3PR09MB78750B229CD254268665AFF5CE709%40BY3PR09MB7875.namprd09.prod.outlook.com.